0.0.14: DataLoader now support video and streaming

- Added `Hub` for resource management
- Updated `DataLoader` to support video and streaming
- Updated `CI`
- Replaced `println!` with `tracing` for logging

.github/workflows/rust-ci.yml

@@ -1,78 +1,203 @@
-name: Rust
+name: Rust-CI
 
 on:
   push:
-    branches: [ "main", "dev" ]
+    branches: [ "main", "dev", "develop", "x", "xy" , "xyz" ]
   pull_request:
     branches: [ "main" ]
 
+env:
+  CARGO_TERM_COLOR: always
+
+
 jobs:
-  check:
-    name: Check
-    runs-on: ${{ matrix.os }}
+  build-on-linux:
+    name: build / linux / ffmpeg ${{ matrix.ffmpeg_version }}
+    runs-on: ubuntu-latest
+    container: jrottenberg/ffmpeg:${{ matrix.ffmpeg_version }}-ubuntu
+
     strategy:
       matrix:
-        os: [ubuntu-latest, macOS-latest, windows-latest]
-        rust: [stable]
-    steps:
-      - uses: actions/checkout@v2
-      - uses: actions-rs/toolchain@v1
-        with:
-          profile: minimal
-          toolchain: ${{ matrix.rust }}
-          override: true
-      - uses: actions-rs/cargo@v1
-        with:
-          command: check
-          args: --workspace --examples
+        ffmpeg_version: ["4.3", "4.4", "5.0", "5.1", "6.0", "6.1", "7.0"]
+      fail-fast: false
+
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v3
+
+      - name: Install dependencies
+        run: |
+          apt update
+          apt install -y --no-install-recommends clang curl pkg-config
+
+      - name: Setup Rust
+        uses: dtolnay/rust-toolchain@v1
+        with:
+          toolchain: stable
+    
+      - name: Build
+        run: cargo build
+
+  build-on-macos:
+    name: build / macos / ffmpeg latest
+    runs-on: macos-latest
+
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v3
+
+      - name: Install dependencies
+        run: |
+          brew install ffmpeg pkg-config
+
+      - name: Setup Rust
+        uses: dtolnay/rust-toolchain@v1
+        with:
+          toolchain: stable
+    
+      - name: Build
+        run: cargo build
+
+
+  build-on-windows:
+    name: build / windows / ffmpeg latest
+    runs-on: windows-latest
+
+    env:
+      FFMPEG_DOWNLOAD_URL: https://www.gyan.dev/ffmpeg/builds/ffmpeg-release-full-shared.7z
+
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v3
+
+      - name: Install dependencies
+        run: |
+          $VCINSTALLDIR = $(& "${env:ProgramFiles(x86)}\Microsoft Visual Studio\Installer\vswhere.exe" -latest -property installationPath)
+          Add-Content $env:GITHUB_ENV "LIBCLANG_PATH=${VCINSTALLDIR}\VC\Tools\LLVM\x64\bin`n"
+          Invoke-WebRequest "${env:FFMPEG_DOWNLOAD_URL}" -OutFile ffmpeg-release-full-shared.7z
+          7z x ffmpeg-release-full-shared.7z
+          mkdir ffmpeg
+          mv ffmpeg-*/* ffmpeg/
+          Add-Content $env:GITHUB_ENV "FFMPEG_DIR=${pwd}\ffmpeg`n"
+          Add-Content $env:GITHUB_PATH "${pwd}\ffmpeg\bin`n"
+
+      - name: Setup Rust
+        uses: dtolnay/rust-toolchain@v1
+        with:
+          toolchain: stable
+    
+      - name: Build
+        run: cargo build
+
+
+  test-on-linux:
+    name: test / linux / ffmpeg ${{ matrix.ffmpeg_version }}
+    runs-on: ubuntu-latest
+    container: jrottenberg/ffmpeg:${{ matrix.ffmpeg_version }}-ubuntu
 
-  test:
-    name: Test
-    runs-on: ${{ matrix.os }}
     strategy:
       matrix:
-        os: [ubuntu-latest, macOS-latest, windows-latest]
-        rust: [stable]
-    steps:
-      - uses: actions/checkout@v2
-      - uses: actions-rs/toolchain@v1
-        with:
-          profile: minimal
-          toolchain: ${{ matrix.rust }}
-          override: true
-      - uses: actions-rs/cargo@v1
-        with:
-          command: test
-          args: --workspace --examples
+        ffmpeg_version: ["4.3", "4.4", "5.0", "5.1", "6.0", "6.1", "7.0"]
+      fail-fast: false
 
-  fmt:
-    name: Rustfmt
-    runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v2
-      - uses: actions-rs/toolchain@v1
-        with:
-          profile: minimal
-          toolchain: stable
-          override: true
-      - run: rustup component add rustfmt
-      - uses: actions-rs/cargo@v1
-        with:
-          command: fmt
-          args: --all -- --check
+      - name: Checkout
+        uses: actions/checkout@v3
 
-  clippy:
-    name: Clippy
-    runs-on: ubuntu-latest
-    steps:
-      - uses: actions/checkout@v2
-      - uses: actions-rs/toolchain@v1
+      - name: Install dependencies
+        run: |
+          apt update
+          apt install -y --no-install-recommends clang curl pkg-config
+
+      - name: Setup Rust
+        uses: dtolnay/rust-toolchain@v1
         with:
-          profile: minimal
           toolchain: stable
-          override: true
-      - run: rustup component add clippy
-      - uses: actions-rs/cargo@v1
+
+      - name: Run Tests with All Features
+        run: cargo test --all-features
+
+      - name: Run Tests in Release Mode
+        run: cargo test --release
+
+  test-on-macos:
+    name: test / macos / ffmpeg latest 
+    runs-on: macos-latest
+
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v3
+
+      - name: Install dependencies
+        run: |
+          brew install ffmpeg pkg-config
+
+      - name: Setup Rust
+        uses: dtolnay/rust-toolchain@v1
         with:
-          command: clippy
-          args: --workspace --tests --examples --all-targets --all-features -- -Dwarnings
+          toolchain: stable
+    
+      - name: Run Tests with All Features
+        run: cargo test --all-features
+
+      - name: Run Tests in Release Mode
+        run: cargo test --release
+
+  test-on-windows:
+    name: test / windows / ffmpeg latest
+    runs-on: windows-latest
+
+    env:
+      FFMPEG_DOWNLOAD_URL: https://www.gyan.dev/ffmpeg/builds/ffmpeg-release-full-shared.7z
+
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v3
+
+      - name: Install dependencies
+        run: |
+          $VCINSTALLDIR = $(& "${env:ProgramFiles(x86)}\Microsoft Visual Studio\Installer\vswhere.exe" -latest -property installationPath)
+          Add-Content $env:GITHUB_ENV "LIBCLANG_PATH=${VCINSTALLDIR}\VC\Tools\LLVM\x64\bin`n"
+          Invoke-WebRequest "${env:FFMPEG_DOWNLOAD_URL}" -OutFile ffmpeg-release-full-shared.7z
+          7z x ffmpeg-release-full-shared.7z
+          mkdir ffmpeg
+          mv ffmpeg-*/* ffmpeg/
+          Add-Content $env:GITHUB_ENV "FFMPEG_DIR=${pwd}\ffmpeg`n"
+          Add-Content $env:GITHUB_PATH "${pwd}\ffmpeg\bin`n"
+
+      - name: Setup Rust
+        uses: dtolnay/rust-toolchain@v1
+        with:
+          toolchain: stable
+    
+      - name: Run Tests with All Features
+        run: cargo test --all-features
+
+      - name: Run Tests in Release Mode
+        run: cargo test --release
+
+
+  lints:
+    runs-on: ubuntu-latest
+    container: jrottenberg/ffmpeg:6-ubuntu
+
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v3
+
+      - name: Install dependencies
+        run: |
+          apt update
+          apt install -y --no-install-recommends clang curl pkg-config
+
+      - name: Setup Rust
+        uses: dtolnay/rust-toolchain@v1
+        with:
+          toolchain: stable
+          components: rustfmt, clippy
+    
+      - name: Rustfmt
+        run: cargo fmt --all -- --check
+
+      - name: Clippy
+        run: cargo clippy --all --all-targets --all-features -- -D warnings

Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "usls"
-version = "0.0.13"
+version = "0.0.14"
 edition = "2021"
 description = "A Rust library integrated with ONNXRuntime, providing a collection of ML models."
 repository = "https://github.com/jamjamjon/usls"
@@ -22,7 +22,7 @@ dirs = { version = "5.0.1" }
 ureq = { version = "2.9.1", default-features = true, features = [
     "socks-proxy",
 ] }
-walkdir = { version = "2.5.0" }
+walkdir = { version = "2.5.0" }  # TODO: remove
 tokenizers = { version = "0.15.2" }
 rayon = "1.10.0"
 indicatif = "0.17.8"
@@ -32,6 +32,13 @@ ab_glyph = "0.2.23"
 geo = "0.28.0"
 prost = "0.12.4"
 fast_image_resize = { version = "4.2.1", features = ["image"]}
+serde = { version = "1.0", features = ["derive"] }
+serde_json = "1.0"
+tempfile = "3.12.0"
+video-rs = { version = "0.9.0", features = ["ndarray"] }
+natord = "1.0.9"
+tracing = "0.1.40"
+tracing-subscriber = "0.3.18"
 
 
 [features]

README.md

@@ -98,7 +98,7 @@ You have two options to link the ONNXRuntime library
 
 
 
-## 🎈 Quick Start
+## 🎈 Demo
 
 ```Shell
 cargo run -r --example yolo   # blip, clip, yolop, svtr, db, ...
@@ -106,7 +106,7 @@ cargo run -r --example yolo   # blip, clip, yolop, svtr, db, ...
 
 ## 🥂 Integrate Into Your Own Project
 
-Add `usls` as a dependency to your project's `Cargo.toml`
+- #### Add `usls` as a dependency to your project's `Cargo.toml`
     ```Shell
     cargo add usls
     ```
@@ -117,5 +117,74 @@ Or use a specific commit:
     usls = { git = "https://github.com/jamjamjon/usls", rev = "commit-sha" }
     ```
     
+- #### Follow the pipeline
+    - Build model with the provided `models` and `Options`
+    - Load images, video and stream with `DataLoader`
+    - Do inference
+    - Annotate inference results with `Annotator`
+    - Retrieve inference results from `Vec<Y>`
+           
+      ```rust
+        use usls::{models::YOLO, Annotator, DataLoader, Nms, Options, Vision, YOLOTask, YOLOVersion};
+    
+        fn main() -> anyhow::Result<()> {
+            // Build model with Options
+            let options = Options::new()
+                .with_trt(0)
+                .with_model("yolo/v8-m-dyn.onnx")?
+                .with_yolo_version(YOLOVersion::V8) // YOLOVersion: V5, V6, V7, V8, V9, V10, RTDETR
+                .with_yolo_task(YOLOTask::Detect) // YOLOTask: Classify, Detect, Pose, Segment, Obb
+                .with_i00((1, 2, 4).into())
+                .with_i02((0, 640, 640).into())
+                .with_i03((0, 640, 640).into())
+                .with_confs(&[0.2]);
+            let mut model = YOLO::new(options)?;
+        
+            // Build DataLoader to load image(s), video, stream
+            let dl = DataLoader::new(
+                // "./assets/bus.jpg", // local image
+                // "images/bus.jpg",  // remote image
+                // "../images-folder",  // local images (from folder)
+                // "../demo.mp4",  // local video
+                // "http://commondatastorage.googleapis.com/gtv-videos-bucket/sample/BigBuckBunny.mp4",  // remote video
+                "rtsp://admin:kkasd1234@192.168.2.217:554/h264/ch1/",  // stream
+            )?
+            .with_batch(2)  // iterate with batch_size = 2
+            .build()?;
+        
+            // Build annotator
+            let annotator = Annotator::new()
+                .with_bboxes_thickness(4)
+                .with_saveout("YOLO-DataLoader");
+        
+            // Run and annotate results
+            for (xs, _) in dl {
+                let ys = model.forward(&xs, false)?;
+                annotator.annotate(&xs, &ys);
+      
+                // Retrieve inference results
+                for y in ys {
+                    // bboxes
+                    if let Some(bboxes) = y.bboxes() {
+                        for bbox in bboxes {
+                            println!(
+                                "Bbox: {}, {}, {}, {}, {}, {}",
+                                bbox.xmin(),
+                                bbox.ymin(),
+                                bbox.xmax(),
+                                bbox.ymax(),
+                                bbox.confidence(),
+                                bbox.id(),
+                            );
+                        }
+                    }
+                }
+            }
+        
+            Ok(())
+        }
+      ```
+
+
 ## 📌 License
 This project is licensed under [LICENSE](LICENSE).

benches/yolo.rs

@@ -1,7 +1,7 @@
 use anyhow::Result;
 use criterion::{black_box, criterion_group, criterion_main, Criterion};
 
-use usls::{models::YOLO, DataLoader, Options, Vision, YOLOTask, YOLOVersion, COCO_KEYPOINTS_17};
+use usls::{models::YOLO, DataLoader, Options, Vision, YOLOTask, YOLOVersion};
 
 enum Stage {
     Pre,
@@ -52,18 +52,17 @@ pub fn benchmark_cuda(c: &mut Criterion, h: isize, w: isize) -> Result<()> {
     let options = Options::default()
         .with_yolo_version(YOLOVersion::V8) // YOLOVersion: V5, V6, V7, V8, V9, V10, RTDETR
         .with_yolo_task(YOLOTask::Detect) // YOLOTask: Classify, Detect, Pose, Segment, Obb
-        .with_model("yolov8m-dyn.onnx")?
+        .with_model("yolo/v8-m-dyn.onnx")?
         .with_cuda(0)
         // .with_cpu()
         .with_dry_run(0)
         .with_i00((1, 1, 4).into())
         .with_i02((320, h, 1280).into())
         .with_i03((320, w, 1280).into())
-        .with_confs(&[0.2, 0.15]) // class_0: 0.4, others: 0.15
-        .with_names2(&COCO_KEYPOINTS_17);
+        .with_confs(&[0.2, 0.15]);
     let mut model = YOLO::new(options)?;
 
-    let xs = vec![DataLoader::try_read("./assets/bus.jpg")?];
+    let xs = [DataLoader::try_read("./assets/bus.jpg")?];
 
     group.bench_function("pre-process", |b| {
         b.iter_custom(|n| yolo_stage_bench(&mut model, &xs, Stage::Pre, n))

examples/blip/README.md

@@ -6,17 +6,11 @@ This demo shows how to use [BLIP](https://arxiv.org/abs/2201.12086) to do condit
 cargo run -r --example blip
 ```
 
-## BLIP ONNX Model
-
-- [blip-visual-base](https://github.com/jamjamjon/assets/releases/download/v0.0.1/blip-visual-base.onnx)  
-- [blip-textual-base](https://github.com/jamjamjon/assets/releases/download/v0.0.1/blip-textual-base.onnx)
-
-
 ## Results
 
 ```shell
-[Unconditional image captioning]: a group of people walking around a bus
-[Conditional image captioning]: three man walking in front of a bus
+[Unconditional]: a group of people walking around a bus
+[Conditional]: three man walking in front of a bus
 Some(["three man walking in front of a bus"])
 ```
 

examples/blip/main.rs

@@ -3,14 +3,14 @@ use usls::{models::Blip, DataLoader, Options};
 fn main() -> Result<(), Box<dyn std::error::Error>> {
     // visual
     let options_visual = Options::default()
-        .with_model("blip-visual-base.onnx")?
+        .with_model("blip/visual-base.onnx")?
         .with_i00((1, 1, 4).into())
         .with_profile(false);
 
     // textual
     let options_textual = Options::default()
-        .with_model("blip-textual-base.onnx")?
-        // .with_tokenizer("tokenizer-blip.json")?
+        .with_model("blip/textual-base.onnx")?
+        // .with_tokenizer("blip/tokenizer.json")?
         .with_i00((1, 1, 4).into()) // input_id: batch
         .with_i01((1, 1, 4).into()) // input_id: seq_len
         .with_i10((1, 1, 4).into()) // attention_mask: batch
@@ -23,7 +23,7 @@ fn main() -> Result<(), Box<dyn std::error::Error>> {
     let mut model = Blip::new(options_visual, options_textual)?;
 
     // image caption (this demo use batch_size=1)
-    let xs = vec![DataLoader::try_read("./assets/bus.jpg")?];
+    let xs = [DataLoader::try_read("images/bus.jpg")?];
     let image_embeddings = model.encode_images(&xs)?;
     let _y = model.caption(&image_embeddings, None, true)?; // unconditional
     let y = model.caption(&image_embeddings, Some("three man"), true)?; // conditional

examples/clip/README.md

@@ -6,12 +6,6 @@ This demo showcases how to use [CLIP](https://github.com/openai/CLIP) to compute
 cargo run -r --example clip
 ```
 
-## CLIP ONNX Model
-
-- [clip-b32-visual](https://github.com/jamjamjon/assets/releases/download/v0.0.1/clip-b32-visual.onnx)  
-- [clip-b32-textual](https://github.com/jamjamjon/assets/releases/download/v0.0.1/clip-b32-textual.onnx)
-
-
 ## Results
 
 ```shell

examples/clip/main.rs

@@ -3,14 +3,14 @@ use usls::{models::Clip, DataLoader, Options};
 fn main() -> Result<(), Box<dyn std::error::Error>> {
     // visual
     let options_visual = Options::default()
-        .with_model("clip-b32-visual-dyn.onnx")?
+        .with_model("clip/visual-base-dyn.onnx")?
         .with_i00((1, 1, 4).into())
         .with_profile(false);
 
     // textual
     let options_textual = Options::default()
-        .with_model("clip-b32-textual-dyn.onnx")?
-        // .with_tokenizer("tokenizer-clip.json")?
+        .with_model("clip/textual-base-dyn.onnx")?
+        .with_tokenizer("clip/tokenizer.json")?
         .with_i00((1, 1, 4).into())
         .with_profile(false);
 
@@ -30,9 +30,7 @@ fn main() -> Result<(), Box<dyn std::error::Error>> {
     let feats_text = model.encode_texts(&texts)?; // [n, ndim]
 
     // load image
-    let dl = DataLoader::default()
-        .with_batch(model.batch_visual())
-        .load("./examples/clip/images")?;
+    let dl = DataLoader::new("./examples/clip/images")?.build()?;
 
     // loop
     for (images, paths) in dl {

examples/dataloader/main.rs

@@ -0,0 +1,49 @@
+use usls::{models::YOLO, Annotator, DataLoader, Options, Vision, YOLOTask, YOLOVersion};
+
+fn main() -> anyhow::Result<()> {
+    tracing_subscriber::fmt()
+        .with_max_level(tracing::Level::ERROR)
+        .init();
+
+    let options = Options::new()
+        .with_cuda(0)
+        .with_model("yolo/v8-m-dyn.onnx")?
+        .with_yolo_version(YOLOVersion::V8)
+        .with_yolo_task(YOLOTask::Detect)
+        .with_i00((1, 1, 4).into())
+        .with_i02((0, 640, 640).into())
+        .with_i03((0, 640, 640).into())
+        .with_confs(&[0.2]);
+    let mut model = YOLO::new(options)?;
+
+    // build dataloader
+    let dl = DataLoader::new(
+        "./assets/bus.jpg", // local image
+                            // "images/bus.jpg",  // remote image
+                            // "../images", // image folder
+                            // "../demo.mp4",   // local video
+                            // "http://commondatastorage.googleapis.com/gtv-videos-bucket/sample/BigBuckBunny.mp4",  // remote video
+                            // "rtsp://admin:xyz@192.168.2.217:554/h265/ch1/",  // rtsp h264 stream
+    )?
+    .with_batch(1)
+    .with_progress_bar(true)
+    .with_bound(100)
+    .build()?;
+
+    // // build annotator
+    let annotator = Annotator::new()
+        .with_bboxes_thickness(4)
+        .with_saveout("YOLO-DataLoader");
+
+    // run
+    for (xs, _) in dl {
+        // std::thread::sleep(std::time::Duration::from_millis(1000));
+        let ys = model.forward(&xs, false)?;
+        annotator.annotate(&xs, &ys);
+    }
+
+    // images -> video
+    // DataLoader::is2v("runs/YOLO-DataLoader", &["runs", "is2v"], 24)?;
+
+    Ok(())
+}

examples/db/README.md

@@ -4,11 +4,6 @@
 cargo run -r --example db
 ```
 
-## ONNX Model
-
-- [ppocr-v3-db-dyn](https://github.com/jamjamjon/assets/releases/download/v0.0.1/ppocr-v3-db-dyn.onnx)  
-- [ppocr-v4-db-dyn](https://github.com/jamjamjon/assets/releases/download/v0.0.1/ppocr-v4-db-dyn.onnx)
-
 ### Speed test
 
 | Model           | Image size | TensorRT<br />f16<br />batch=1<br />(ms) | TensorRT<br />f32<br />batch=1<br />(ms) | CUDA<br />f32<br />batch=1<br />(ms) |
@@ -20,4 +15,5 @@ cargo run -r --example db
 
 ## Results
 
-![](./demo.png)
+![](https://github.com/jamjamjon/assets/releases/download/db/demo-paper.png)
+![](https://github.com/jamjamjon/assets/releases/download/db/demo-sign.png)

examples/db/main.rs

@@ -10,14 +10,14 @@ fn main() -> Result<(), Box<dyn std::error::Error>> {
         .with_confs(&[0.4])
         .with_min_width(5.0)
         .with_min_height(12.0)
-        .with_model("ppocr-v4-db-dyn.onnx")?;
+        .with_model("db/ppocr-v4-db-dyn.onnx")?;
 
     let mut model = DB::new(options)?;
 
     // load image
-    let x = vec![
-        DataLoader::try_read("./assets/db.png")?,
-        DataLoader::try_read("./assets/2.jpg")?,
+    let x = [
+        DataLoader::try_read("images/db.png")?,
+        DataLoader::try_read("images/street.jpg")?,
     ];
 
     // run

examples/depth-anything/README.md

@@ -4,14 +4,7 @@
 cargo run -r --example depth-anything
 ```
 
-## ONNX Model
-
-- [depth-anything-s-dyn](https://github.com/jamjamjon/assets/releases/download/v0.0.1/depth-anything-s-dyn.onnx)  
-- [depth-anything-b-dyn](https://github.com/jamjamjon/assets/releases/download/v0.0.1/depth-anything-b-dyn.onnx)
-- [depth-anything-l-dyn](https://github.com/jamjamjon/assets/releases/download/v0.0.1/depth-anything-l-dyn.onnx)
-- [depth-anything-v2-s](https://github.com/jamjamjon/assets/releases/download/v0.0.1/depth-anything-v2-s.onnx)  
-
 
 ## Results
 
-![](./demo.png)
+![](https://github.com/jamjamjon/assets/releases/download/depth-anything/demo.png)

examples/depth-anything/main.rs

@@ -3,15 +3,15 @@ use usls::{models::DepthAnything, Annotator, DataLoader, Options};
 fn main() -> Result<(), Box<dyn std::error::Error>> {
     // options
     let options = Options::default()
-        // .with_model("depth-anything-s-dyn.onnx")?
-        .with_model("depth-anything-v2-s.onnx")?
+        // .with_model("depth-anything/v1-s-dyn.onnx")?
+        .with_model("depth-anything/v2-s.onnx")?
         .with_i00((1, 1, 8).into())
         .with_i02((384, 512, 1024).into())
         .with_i03((384, 512, 1024).into());
     let mut model = DepthAnything::new(options)?;
 
     // load
-    let x = [DataLoader::try_read("./assets/2.jpg")?];
+    let x = [DataLoader::try_read("images/2.jpg")?];
 
     // run
     let y = model.run(&x)?;

examples/dinov2/README.md

@@ -5,22 +5,3 @@ This demo showcases how to use `DINOv2` to compute image similarity, applicable
 ```shell
 cargo run -r --example dinov2
 ```
-
-## Donwload DINOv2 ONNX Model
-
-- [dinov2-s14](https://github.com/jamjamjon/assets/releases/download/v0.0.1/dinov2-s14.onnx)
-- [dinov2-s14-dyn](https://github.com/jamjamjon/assets/releases/download/v0.0.1/dinov2-s14-dyn.onnx)
-- [dinov2-s14-dyn-f16](https://github.com/jamjamjon/assets/releases/download/v0.0.1/dinov2-s14-dyn-f16.onnx)
-
-- [dinov2-b14](https://github.com/jamjamjon/assets/releases/download/v0.0.1/dinov2-b14.onnx)
-- [dinov2-b14-dyn](https://github.com/jamjamjon/assets/releases/download/v0.0.1/dinov2-b14-dyn.onnx)
-
-
-
-## Results
-
-```shell
-Top-1  0.0000000 /home/qweasd/Desktop/usls/examples/dinov2/images/bus.jpg
-Top-2  1.9059424 /home/qweasd/Desktop/usls/examples/dinov2/images/1.jpg
-Top-3  1.9736203 /home/qweasd/Desktop/usls/examples/dinov2/images/2.jpg
-```

examples/dinov2/main.rs

@@ -3,15 +3,16 @@ use usls::{models::Dinov2, DataLoader, Options};
 fn main() -> Result<(), Box<dyn std::error::Error>> {
     // build model
     let options = Options::default()
-        .with_model("dinov2-s14-dyn-f16.onnx")?
+        .with_model("dinov2/s-dyn.onnx")?
         .with_i00((1, 1, 1).into())
         .with_i02((224, 224, 224).into())
         .with_i03((224, 224, 224).into());
     let mut model = Dinov2::new(options)?;
-    let x = vec![DataLoader::try_read("./examples/dinov2/images/1.jpg")?];
+    let x = [DataLoader::try_read("images/bus.jpg")?];
     let y = model.run(&x)?;
     println!("{y:?}");
 
+    // TODO:
     // query from vector
     // let ys = model.query_from_vec(
     //     "./assets/bus.jpg",

examples/grounding-dino/README.md

@@ -0,0 +1,10 @@
+## Quick Start
+
+```shell
+cargo run -r --example grounding-dino
+```
+
+
+## Results
+
+![](https://github.com/jamjamjon/assets/releases/download/grounding-dino/demo.png)

examples/grounding-dino/main.rs

@@ -16,14 +16,15 @@ fn main() -> Result<(), Box<dyn std::error::Error>> {
         .with_i50((1, 1, 4).into())
         .with_i51((256, 256, 512).into())
         .with_i52((256, 256, 512).into())
-        .with_model("groundingdino-swint-ogc-dyn-u8.onnx")? // TODO: current onnx model does not support bs > 1
-        // .with_model("groundingdino-swint-ogc-dyn-f32.onnx")?
+        .with_model("grounding-dino/swint-ogc-dyn-u8.onnx")? // TODO: current onnx model does not support bs > 1
+        // .with_model("grounding-dino/swint-ogc-dyn-f32.onnx")?
+        .with_tokenizer("grounding-dino/tokenizer.json")?
         .with_confs(&[0.2])
         .with_profile(false);
     let mut model = GroundingDINO::new(opts)?;
 
     // Load images and set class names
-    let x = [DataLoader::try_read("./assets/bus.jpg")?];
+    let x = [DataLoader::try_read("images/bus.jpg")?];
     let texts = [
         "person", "hand", "shoes", "bus", "dog", "cat", "sign", "tie", "monitor", "window",
         "glasses", "tree", "head",

examples/modnet/README.md

@@ -4,12 +4,7 @@
 cargo run -r --example modnet
 ```
 
-## ONNX Model
-
-- [modnet-dyn](https://github.com/jamjamjon/assets/releases/download/v0.0.1/modnet-dyn.onnx)  
-
-
 
 ## Results
 
-![](./demo.png)
+![](https://github.com/jamjamjon/assets/releases/download/modnet/demo.png)

examples/modnet/main.rs

@@ -3,14 +3,14 @@ use usls::{models::MODNet, Annotator, DataLoader, Options};
 fn main() -> Result<(), Box<dyn std::error::Error>> {
     // build model
     let options = Options::default()
-        .with_model("modnet-dyn.onnx")?
+        .with_model("modnet/dyn-f32.onnx")?
         .with_i00((1, 1, 4).into())
         .with_i02((416, 512, 800).into())
         .with_i03((416, 512, 800).into());
     let mut model = MODNet::new(options)?;
 
     // load image
-    let x = vec![DataLoader::try_read("./assets/liuyifei.png")?];
+    let x = [DataLoader::try_read("images/liuyifei.png")?];
 
     // run
     let y = model.run(&x)?;

examples/rtmo/README.md

@@ -4,15 +4,7 @@
 cargo run -r --example rtmo
 ```
  
-## ONNX Model
-
-- [rtmo-s-dyn model](https://github.com/jamjamjon/assets/releases/download/v0.0.1/rtmo-s-dyn.onnx)  
-- [rtmo-m-dyn model](https://github.com/jamjamjon/assets/releases/download/v0.0.1/rtmo-m-dyn.onnx)  
-- [rtmo-l-dyn model](https://github.com/jamjamjon/assets/releases/download/v0.0.1/rtmo-l-dyn.onnx)  
-- [rtmo-s-dyn-f16 model](https://github.com/jamjamjon/assets/releases/download/v0.0.1/rtmo-s-dyn-f16.onnx)  
-- [rtmo-m-dyn-f16 model](https://github.com/jamjamjon/assets/releases/download/v0.0.1/rtmo-m-dyn-f16.onnx)  
-- [rtmo-l-dyn-f16 model](https://github.com/jamjamjon/assets/releases/download/v0.0.1/rtmo-l-dyn-f16.onnx)  
 
 ## Results
 
-![](./demo.png)
+![](https://github.com/jamjamjon/assets/releases/download/rtmo/demo.png)

examples/rtmo/main.rs

@@ -3,7 +3,7 @@ use usls::{models::RTMO, Annotator, DataLoader, Options, COCO_SKELETONS_16};
 fn main() -> Result<(), Box<dyn std::error::Error>> {
     // build model
     let options = Options::default()
-        .with_model("rtmo-s-dyn.onnx")?
+        .with_model("rtmo/s-dyn.onnx")?
         .with_i00((1, 1, 8).into())
         .with_nk(17)
         .with_confs(&[0.3])
@@ -11,7 +11,7 @@ fn main() -> Result<(), Box<dyn std::error::Error>> {
     let mut model = RTMO::new(options)?;
 
     // load image
-    let x = vec![DataLoader::try_read("./assets/bus.jpg")?];
+    let x = [DataLoader::try_read("images/bus.jpg")?];
 
     // run
     let y = model.run(&x)?;

examples/sam/README.md

@@ -18,4 +18,5 @@ cargo run -r --example sam -- --kind sam-hq
 
 ## Results
 
-![](./demo.png)
+![](https://github.com/jamjamjon/assets/releases/download/sam/demo-car.png)
+![](https://github.com/jamjamjon/assets/releases/download/sam/demo-dog.png)

examples/sam/main.rs

@@ -25,63 +25,64 @@ fn main() -> Result<(), Box<dyn std::error::Error>> {
     let (options_encoder, options_decoder, saveout) = match args.kind {
         SamKind::Sam => {
             let options_encoder = Options::default()
-                // .with_model("sam-vit-b-encoder.onnx")?;
-                .with_model("sam-vit-b-encoder-u8.onnx")?;
+                // .with_model("sam/sam-vit-b-encoder.onnx")?;
+                .with_model("sam/sam-vit-b-encoder-u8.onnx")?;
 
             let options_decoder = Options::default()
                 .with_i00((1, 1, 1).into())
                 .with_i11((1, 1, 1).into())
                 .with_i21((1, 1, 1).into())
                 .with_sam_kind(SamKind::Sam)
-                // .with_model("sam-vit-b-decoder.onnx")?;
-                // .with_model("sam-vit-b-decoder-singlemask.onnx")?;
-                .with_model("sam-vit-b-decoder-u8.onnx")?;
+                // .with_model("sam/sam-vit-b-decoder.onnx")?;
+                // .with_model("sam/sam-vit-b-decoder-singlemask.onnx")?;
+                .with_model("sam/sam-vit-b-decoder-u8.onnx")?;
             (options_encoder, options_decoder, "SAM")
         }
         SamKind::Sam2 => {
             let options_encoder = Options::default()
-                // .with_model("sam2-hiera-tiny-encoder.onnx")?;
-                // .with_model("sam2-hiera-small-encoder.onnx")?;
-                .with_model("sam2-hiera-base-plus-encoder.onnx")?;
+                // .with_model("sam/sam2-hiera-tiny-encoder.onnx")?;
+                // .with_model("sam/sam2-hiera-small-encoder.onnx")?;
+                .with_model("sam/sam2-hiera-base-plus-encoder.onnx")?;
             let options_decoder = Options::default()
                 .with_i31((1, 1, 1).into())
                 .with_i41((1, 1, 1).into())
                 .with_sam_kind(SamKind::Sam2)
-                // .with_model("sam2-hiera-tiny-decoder.onnx")?;
-                // .with_model("sam2-hiera-small-decoder.onnx")?;
-                .with_model("sam2-hiera-base-plus-decoder.onnx")?;
+                // .with_model("sam/sam2-hiera-tiny-decoder.onnx")?;
+                // .with_model("sam/sam2-hiera-small-decoder.onnx")?;
+                .with_model("sam/sam2-hiera-base-plus-decoder.onnx")?;
             (options_encoder, options_decoder, "SAM2")
         }
         SamKind::MobileSam => {
-            let options_encoder = Options::default().with_model("mobile-sam-vit-t-encoder.onnx")?;
+            let options_encoder =
+                Options::default().with_model("sam/mobile-sam-vit-t-encoder.onnx")?;
 
             let options_decoder = Options::default()
                 .with_i00((1, 1, 1).into())
                 .with_i11((1, 1, 1).into())
                 .with_i21((1, 1, 1).into())
                 .with_sam_kind(SamKind::MobileSam)
-                .with_model("mobile-sam-vit-t-decoder.onnx")?;
+                .with_model("sam/mobile-sam-vit-t-decoder.onnx")?;
             (options_encoder, options_decoder, "Mobile-SAM")
         }
         SamKind::SamHq => {
-            let options_encoder = Options::default().with_model("sam-hq-vit-t-encoder.onnx")?;
+            let options_encoder = Options::default().with_model("sam/sam-hq-vit-t-encoder.onnx")?;
 
             let options_decoder = Options::default()
                 .with_i00((1, 1, 1).into())
                 .with_i21((1, 1, 1).into())
                 .with_i31((1, 1, 1).into())
                 .with_sam_kind(SamKind::SamHq)
-                .with_model("sam-hq-vit-t-decoder.onnx")?;
+                .with_model("sam/sam-hq-vit-t-decoder.onnx")?;
             (options_encoder, options_decoder, "SAM-HQ")
         }
         SamKind::EdgeSam => {
-            let options_encoder = Options::default().with_model("edge-sam-3x-encoder.onnx")?;
+            let options_encoder = Options::default().with_model("sam/edge-sam-3x-encoder.onnx")?;
             let options_decoder = Options::default()
                 .with_i00((1, 1, 1).into())
                 .with_i11((1, 1, 1).into())
                 .with_i21((1, 1, 1).into())
                 .with_sam_kind(SamKind::EdgeSam)
-                .with_model("edge-sam-3x-decoder.onnx")?;
+                .with_model("sam/edge-sam-3x-decoder.onnx")?;
             (options_encoder, options_decoder, "Edge-SAM")
         }
     };
@@ -100,8 +101,8 @@ fn main() -> Result<(), Box<dyn std::error::Error>> {
 
     // Load image
     let xs = [
-        DataLoader::try_read("./assets/truck.jpg")?,
-        // DataLoader::try_read("./assets/dog.jpg")?,
+        DataLoader::try_read("images/truck.jpg")?,
+        // DataLoader::try_read("images/dog.jpg")?,
     ];
 
     // Build annotator

examples/sapiens/README.md

@@ -0,0 +1,10 @@
+## Quick Start
+
+```shell
+cargo run -r --example sapiens
+```
+
+
+## Results
+
+![](https://github.com/jamjamjon/assets/releases/download/sapiens/demo.png)

examples/sapiens/main.rs

@@ -6,7 +6,7 @@ use usls::{
 fn main() -> Result<(), Box<dyn std::error::Error>> {
     // build
     let options = Options::default()
-        .with_model("sapiens-seg-0.3b-dyn.onnx")?
+        .with_model("sapiens/seg-0.3b-dyn.onnx")?
         .with_sapiens_task(SapiensTask::Seg)
         .with_names(&BODY_PARTS_28)
         .with_profile(false)
@@ -14,7 +14,7 @@ fn main() -> Result<(), Box<dyn std::error::Error>> {
     let mut model = Sapiens::new(options)?;
 
     // load
-    let x = [DataLoader::try_read("./assets/paul-george.jpg")?];
+    let x = [DataLoader::try_read("images/paul-george.jpg")?];
 
     // run
     let y = model.run(&x)?;

examples/svtr/README.md

@@ -4,13 +4,6 @@
 cargo run -r --example svtr
 ```
 
-## ONNX Model
-
-- [ppocr-v4-server-svtr-ch-dyn](https://github.com/jamjamjon/assets/releases/download/v0.0.1/ppocr-v4-server-svtr-ch-dyn.onnx)  
-- [ppocr-v4-svtr-ch-dyn](https://github.com/jamjamjon/assets/releases/download/v0.0.1/ppocr-v4-svtr-ch-dyn.onnx)  
-- [ppocr-v3-svtr-ch-dyn](https://github.com/jamjamjon/assets/releases/download/v0.0.1/ppocr-v3-svtr-ch-dyn.onnx)  
-
-
 ### Speed test
 
 | Model                       | Width | TensorRT<br />f16<br />batch=1<br />(ms) | TensorRT<br />f32<br />batch=1<br />(ms) | CUDA<br />f32<br />batch=1<br />(ms) |

examples/svtr/main.rs

@@ -6,14 +6,12 @@ fn main() -> Result<(), Box<dyn std::error::Error>> {
         .with_i00((1, 2, 8).into())
         .with_i03((320, 960, 1600).into())
         .with_confs(&[0.2])
-        .with_vocab("ppocr_rec_vocab.txt")?
-        .with_model("ppocr-v4-svtr-ch-dyn.onnx")?;
+        .with_vocab("svtr/ppocr_rec_vocab.txt")?
+        .with_model("svtr/ppocr-v4-svtr-ch-dyn.onnx")?;
     let mut model = SVTR::new(options)?;
 
     // load images
-    let dl = DataLoader::default()
-        .with_batch(1)
-        .load("./examples/svtr/images")?;
+    let dl = DataLoader::new("./examples/svtr/images")?.build()?;
 
     // run
     for (xs, paths) in dl {

examples/yolo-sam/README.md

@@ -0,0 +1,9 @@
+## Quick Start
+
+```shell
+cargo run -r --example yolo-sam
+```
+
+## Results
+
+![](https://github.com/jamjamjon/assets/releases/download/sam/demo-yolo-sam.png)

examples/yolo-sam/main.rs

@@ -7,20 +7,20 @@ fn main() -> Result<(), Box<dyn std::error::Error>> {
     // build SAM
     let options_encoder = Options::default()
         .with_i00((1, 1, 1).into())
-        .with_model("mobile-sam-vit-t-encoder.onnx")?;
+        .with_model("sam/mobile-sam-vit-t-encoder.onnx")?;
     let options_decoder = Options::default()
         .with_i11((1, 1, 1).into())
         .with_i21((1, 1, 1).into())
         .with_find_contours(true)
         .with_sam_kind(SamKind::Sam)
-        .with_model("mobile-sam-vit-t-decoder.onnx")?;
+        .with_model("sam/mobile-sam-vit-t-decoder.onnx")?;
     let mut sam = SAM::new(options_encoder, options_decoder)?;
 
     // build YOLOv8-Det
     let options_yolo = Options::default()
         .with_yolo_version(YOLOVersion::V8)
         .with_yolo_task(YOLOTask::Detect)
-        .with_model("yolov8m-dyn.onnx")?
+        .with_model("yolo/v8-m-dyn.onnx")?
         .with_cuda(0)
         .with_i00((1, 1, 4).into())
         .with_i02((416, 640, 800).into())
@@ -30,7 +30,7 @@ fn main() -> Result<(), Box<dyn std::error::Error>> {
     let mut yolo = YOLO::new(options_yolo)?;
 
     // load one image
-    let xs = vec![DataLoader::try_read("./assets/dog.jpg")?];
+    let xs = [DataLoader::try_read("images/dog.jpg")?];
 
     // build annotator
     let annotator = Annotator::default()
@@ -38,7 +38,7 @@ fn main() -> Result<(), Box<dyn std::error::Error>> {
         .without_bboxes_name(true)
         .without_bboxes_conf(true)
         .without_mbrs(true)
-        .with_saveout("YOLO+SAM");
+        .with_saveout("YOLO-SAM");
 
     // run & annotate
     let ys_det = yolo.run(&xs)?;

examples/yolo/README.md

@@ -3,19 +3,19 @@
 
 |      Detection     |    Instance Segmentation   |   Pose     | 
 | :---------------: | :------------------------: |:---------------: |
-| <img src='./demos/det.png'  width="300px">  | <img src='./demos/seg.png'  width="300px"> |<img src='./demos/pose.png'  width="300px">  | 
+| <img src='https://github.com/jamjamjon/assets/releases/download/yolo/demo-det.png'  width="300px">  | <img src='https://github.com/jamjamjon/assets/releases/download/yolo/demo-seg.png'  width="300px"> |<img src='https://github.com/jamjamjon/assets/releases/download/yolo/demo-pose.png'  width="300px">  | 
 
 |    Classification   |    Obb   |
 | :------------------------: |:------------------------: |
-|<img src='./demos/cls.png'  width="300px"> |<img src='./demos/obb-2.png'  width="628px">
+|<img src='https://github.com/jamjamjon/assets/releases/download/yolo/demo-cls.png'  width="300px"> |<img src='https://github.com/jamjamjon/assets/releases/download/yolo/demo-obb-2.png'  width="628px">
 
 |    Head Detection   |    Fall Detection   | Trash Detection   |
 | :------------------------: |:------------------------: |:------------------------: |
-|<img src='./demos/head.png'  width="300px"> |<img src='./demos/falldown.png'  width="300px">|<img src='./demos/trash.png'  width="300px">
+|<img src='https://github.com/jamjamjon/assets/releases/download/yolo/demo-head.png'  width="300px"> |<img src='https://github.com/jamjamjon/assets/releases/download/yolo/demo-falldown.png'  width="300px">|<img src='https://github.com/jamjamjon/assets/releases/download/yolo/demo-trash.png'  width="300px">
 
 |    YOLO-World   |    Face Parsing   | FastSAM   |
 | :------------------------: |:------------------------: |:------------------------: |
-|<img src='./demos/yolov8-world.png'  width="300px"> |<img src='./demos/face-parsing.png'  width="300px">|<img src='./demos/fastsam.png'  width="300px">
+|<img src='https://github.com/jamjamjon/assets/releases/download/yolo/demo-yolov8-world.png'  width="300px"> |<img src='https://github.com/jamjamjon/assets/releases/download/yolo/demo-face-parsing.png'  width="300px">|<img src='https://github.com/jamjamjon/assets/releases/download/yolo/demo-fastsam.png'  width="300px">
 
 
 
@@ -104,11 +104,11 @@ let options = Options::default()
 
 |          Model          |    Weights   | Datasets|
 |:---------------------: | :--------------------------: | :-------------------------------: |
-| Face-Landmark Detection |   [yolov8-face-dyn-f16](https://github.com/jamjamjon/assets/releases/download/v0.0.1/yolov8-face-dyn-f16.onnx)         | |
-| Head Detection |   [yolov8-head-f16](https://github.com/jamjamjon/assets/releases/download/v0.0.1/yolov8-head-f16.onnx)         | |
-| Fall Detection |   [yolov8-falldown-f16](https://github.com/jamjamjon/assets/releases/download/v0.0.1/yolov8-falldown-f16.onnx)          | |
-| Trash Detection |   [yolov8-plastic-bag-f16](https://github.com/jamjamjon/assets/releases/download/v0.0.1/yolov8-plastic-bag-f16.onnx)         | |
-| FaceParsing |  [face-parsing-dyn](https://github.com/jamjamjon/assets/releases/download/v0.0.1/face-parsing-dyn.onnx)  | [CelebAMask-HQ](https://github.com/switchablenorms/CelebAMask-HQ/tree/master/face_parsing)<br />[[Processed YOLO labels]](https://github.com/jamjamjon/assets/releases/download/v0.0.1/CelebAMask-HQ-YOLO-Labels.zip)[[Python Script]](https://github.com/jamjamjon/assets/releases/download/v0.0.1/CelebAMask-HQ-YOLO-Labels.zip) |
+| Face-Landmark Detection |   [yolov8-face-dyn-f16](https://github.com/jamjamjon/assets/releases/download/yolo/v8-n-face-dyn-f16.onnx)         | |
+| Head Detection |   [yolov8-head-f16](https://github.com/jamjamjon/assets/releases/download/yolo/v8-head-f16.onnx)         | |
+| Fall Detection |   [yolov8-falldown-f16](https://github.com/jamjamjon/assets/releases/download/yolo/v8-falldown-f16.onnx)          | |
+| Trash Detection |   [yolov8-plastic-bag-f16](https://github.com/jamjamjon/assets/releases/download/yolo/v8-plastic-bag-f16.onnx)         | |
+| FaceParsing |  [yolov8-face-parsing-dyn](https://github.com/jamjamjon/assets/releases/download/yolo/v8-face-parsing-dyn.onnx)  | [CelebAMask-HQ](https://github.com/switchablenorms/CelebAMask-HQ/tree/master/face_parsing)<br />[[Processed YOLO labels]](https://github.com/jamjamjon/assets/releases/download/yolo/CelebAMask-HQ-YOLO-Labels.zip)[[Python Script]](../../scripts/CelebAMask-HQ-To-YOLO-Labels.py) |
 
 
 

examples/yolo/main.rs

@@ -77,15 +77,15 @@ fn main() -> Result<()> {
     let (options, saveout) = match args.ver {
         YOLOVersion::V5 => match args.task {
             YOLOTask::Classify => (
-                options.with_model(&args.model.unwrap_or("yolov5n-cls-dyn.onnx".to_string()))?,
+                options.with_model(&args.model.unwrap_or("yolo/v5-n-cls-dyn.onnx".to_string()))?,
                 "YOLOv5-Classify",
             ),
             YOLOTask::Detect => (
-                options.with_model(&args.model.unwrap_or("yolov5n-dyn.onnx".to_string()))?,
+                options.with_model(&args.model.unwrap_or("yolo/v5-n-dyn.onnx".to_string()))?,
                 "YOLOv5-Detect",
             ),
             YOLOTask::Segment => (
-                options.with_model(&args.model.unwrap_or("yolov5n-seg-dyn.onnx".to_string()))?,
+                options.with_model(&args.model.unwrap_or("yolo/v5-n-seg-dyn.onnx".to_string()))?,
                 "YOLOv5-Segment",
             ),
             t => anyhow::bail!("Task: {t:?} is unsupported for {:?}", args.ver),
@@ -93,7 +93,7 @@ fn main() -> Result<()> {
         YOLOVersion::V6 => match args.task {
             YOLOTask::Detect => (
                 options
-                    .with_model(&args.model.unwrap_or("yolov6n-dyn.onnx".to_string()))?
+                    .with_model(&args.model.unwrap_or("yolo/v6-n-dyn.onnx".to_string()))?
                     .with_nc(args.nc),
                 "YOLOv6-Detect",
             ),
@@ -102,7 +102,7 @@ fn main() -> Result<()> {
         YOLOVersion::V7 => match args.task {
             YOLOTask::Detect => (
                 options
-                    .with_model(&args.model.unwrap_or("yolov7-tiny-dyn.onnx".to_string()))?
+                    .with_model(&args.model.unwrap_or("yolo/v7-tiny-dyn.onnx".to_string()))?
                     .with_nc(args.nc),
                 "YOLOv7-Detect",
             ),
@@ -110,43 +110,43 @@ fn main() -> Result<()> {
         },
         YOLOVersion::V8 => match args.task {
             YOLOTask::Classify => (
-                options.with_model(&args.model.unwrap_or("yolov8m-cls-dyn.onnx".to_string()))?,
+                options.with_model(&args.model.unwrap_or("yolo/v8-m-cls-dyn.onnx".to_string()))?,
                 "YOLOv8-Classify",
             ),
             YOLOTask::Detect => (
-                options.with_model(&args.model.unwrap_or("yolov8m-dyn.onnx".to_string()))?,
+                options.with_model(&args.model.unwrap_or("yolo/v8-m-dyn.onnx".to_string()))?,
                 "YOLOv8-Detect",
             ),
             YOLOTask::Segment => (
-                options.with_model(&args.model.unwrap_or("yolov8m-seg-dyn.onnx".to_string()))?,
+                options.with_model(&args.model.unwrap_or("yolo/v8-m-seg-dyn.onnx".to_string()))?,
                 "YOLOv8-Segment",
             ),
             YOLOTask::Pose => (
-                options.with_model(&args.model.unwrap_or("yolov8m-pose-dyn.onnx".to_string()))?,
+                options.with_model(&args.model.unwrap_or("yolo/v8-m-pose-dyn.onnx".to_string()))?,
                 "YOLOv8-Pose",
             ),
             YOLOTask::Obb => (
-                options.with_model(&args.model.unwrap_or("yolov8m-obb-dyn.onnx".to_string()))?,
+                options.with_model(&args.model.unwrap_or("yolo/v8-m-obb-dyn.onnx".to_string()))?,
                 "YOLOv8-Obb",
             ),
         },
         YOLOVersion::V9 => match args.task {
             YOLOTask::Detect => (
-                options.with_model(&args.model.unwrap_or("yolov9-c-dyn-f16.onnx".to_string()))?,
+                options.with_model(&args.model.unwrap_or("yolo/v9-c-dyn-f16.onnx".to_string()))?,
                 "YOLOv9-Detect",
             ),
             t => anyhow::bail!("Task: {t:?} is unsupported for {:?}", args.ver),
         },
         YOLOVersion::V10 => match args.task {
             YOLOTask::Detect => (
-                options.with_model(&args.model.unwrap_or("yolov10n.onnx".to_string()))?,
+                options.with_model(&args.model.unwrap_or("yolo/v10-n.onnx".to_string()))?,
                 "YOLOv10-Detect",
             ),
             t => anyhow::bail!("Task: {t:?} is unsupported for {:?}", args.ver),
         },
         YOLOVersion::RTDETR => match args.task {
             YOLOTask::Detect => (
-                options.with_model(&args.model.unwrap_or("rtdetr-l-f16.onnx".to_string()))?,
+                options.with_model(&args.model.unwrap_or("yolo/rtdetr-l-f16.onnx".to_string()))?,
                 "RTDETR",
             ),
             t => anyhow::bail!("Task: {t:?} is unsupported for {:?}", args.ver),
@@ -184,9 +184,9 @@ fn main() -> Result<()> {
     let mut model = YOLO::new(options)?;
 
     // build dataloader
-    let dl = DataLoader::default()
+    let dl = DataLoader::new(&args.source)?
         .with_batch(model.batch() as _)
-        .load(args.source)?;
+        .build()?;
 
     // build annotator
     let annotator = Annotator::default()

examples/yolop/README.md

@@ -4,11 +4,6 @@
 cargo run -r --example yolop
 ```
 
-## Pretrained Model
-
-- [yolopv2-dyn-480x800](https://github.com/jamjamjon/assets/releases/download/v0.0.1/yolopv2-dyn-480x800.onnx)  
-- [yolopv2-dyn-736x1280](https://github.com/jamjamjon/assets/releases/download/v0.0.1/yolopv2-dyn-736x1280.onnx)  
-
 ## Results
 
-![](./demo.png)
+![](https://github.com/jamjamjon/assets/releases/download/yolop/demo.png)

examples/yolop/main.rs

@@ -3,13 +3,13 @@ use usls::{models::YOLOPv2, Annotator, DataLoader, Options};
 fn main() -> Result<(), Box<dyn std::error::Error>> {
     // build model
     let options = Options::default()
-        .with_model("yolopv2-dyn-480x800.onnx")?
+        .with_model("yolop/v2-dyn-480x800.onnx")?
         .with_i00((1, 1, 8).into())
         .with_confs(&[0.3]);
     let mut model = YOLOPv2::new(options)?;
 
     // load image
-    let x = vec![DataLoader::try_read("./assets/car.jpg")?];
+    let x = [DataLoader::try_read("images/car.jpg")?];
 
     // run
     let y = model.run(&x)?;

src/core/annotator.rs

@@ -1,5 +1,5 @@
 use crate::{
-    auto_load, colormap256, string_now, Bbox, Keypoint, Mask, Mbr, Polygon, Prob, CHECK_MARK,
+    colormap256, string_now, Bbox, Dir, Hub, Keypoint, Mask, Mbr, Polygon, Prob, CHECK_MARK,
     CROSS_MARK, Y,
 };
 use ab_glyph::{FontVec, PxScale};
@@ -8,11 +8,12 @@ use image::{DynamicImage, GenericImage, Rgba, RgbaImage};
 use imageproc::map::map_colors;
 
 /// Annotator for struct `Y`
-#[derive(Debug)]
+// #[derive(Debug)]
 pub struct Annotator {
     font: FontVec,
     _scale: f32, // Cope with ab_glyph & imageproc=0.24.0
     scale_dy: f32,
+    saveout_base: String,
     saveout: Option<String>,
     decimal_places: usize,
 
@@ -63,11 +64,15 @@ pub struct Annotator {
 impl Default for Annotator {
     fn default() -> Self {
         Self {
-            font: Self::load_font(None).unwrap(),
+            font: match Self::load_font(None) {
+                Ok(x) => x,
+                Err(err) => panic!("Failed to load font: {}", err),
+            },
             _scale: 6.666667,
             scale_dy: 28.,
             polygons_alpha: 179,
             saveout: None,
+            saveout_base: String::from("runs"),
             decimal_places: 4,
             without_bboxes: false,
             without_bboxes_conf: false,
@@ -104,6 +109,10 @@ impl Default for Annotator {
 }
 
 impl Annotator {
+    pub fn new() -> Self {
+        Default::default()
+    }
+
     pub fn with_decimal_places(mut self, x: usize) -> Self {
         self.decimal_places = x;
         self
@@ -304,32 +313,44 @@ impl Annotator {
         self
     }
 
-    pub fn with_saveout(mut self, saveout: &str) -> Self {
-        self.saveout = Some(saveout.to_string());
+    pub fn with_saveout_base(mut self, x: &str) -> Self {
+        self.saveout_base = x.to_string();
         self
     }
 
-    pub fn with_font(mut self, path: &str) -> Self {
-        self.font = Self::load_font(Some(path)).unwrap();
+    pub fn with_saveout(mut self, x: &str) -> Self {
+        self.saveout = Some(x.to_string());
         self
     }
 
-    /// Save annotated images to `runs` folder
-    pub fn save(&self, image: &RgbaImage, saveout: &str) {
-        let mut saveout = std::path::PathBuf::from("runs").join(saveout);
-        if !saveout.exists() {
-            std::fs::create_dir_all(&saveout).unwrap();
-        }
-        saveout.push(string_now("-"));
-        let saveout = format!("{}.png", saveout.to_str().unwrap());
-        match image.save(&saveout) {
-            Err(err) => println!("{} Saving failed: {:?}", CROSS_MARK, err),
-            Ok(_) => println!("{} Annotated image saved to: {}", CHECK_MARK, saveout),
-        }
+    pub fn with_font(mut self, path: &str) -> Result<Self> {
+        self.font = Self::load_font(Some(path))?;
+        Ok(self)
     }
 
-    /// Annotate images
+    /// Create folders for saving annotated results. e.g., `./runs/xxx`
+    pub fn saveout(&self) -> Result<std::path::PathBuf> {
+        let subs = match &self.saveout {
+            Some(x) => vec![self.saveout_base.as_str(), x.as_str()],
+            None => vec![self.saveout_base.as_str()],
+        };
+
+        Dir::Currnet.raw_path_with_subs(&subs)
+    }
+
+    /// Annotate images, and no return
     pub fn annotate(&self, imgs: &[DynamicImage], ys: &[Y]) {
+        let _ = self.plot(imgs, ys);
+    }
+
+    /// Plot images and return plotted images(RGBA8)
+    pub fn plot(&self, imgs: &[DynamicImage], ys: &[Y]) -> Result<Vec<RgbaImage>> {
+        let span = tracing::span!(tracing::Level::INFO, "YOLO-new");
+        let _guard = span.enter();
+
+        let mut vs: Vec<RgbaImage> = Vec::new();
+
+        // annotate
         for (img, y) in imgs.iter().zip(ys.iter()) {
             let mut img_rgba = img.to_rgba8();
 
@@ -374,16 +395,23 @@ impl Annotator {
             }
 
             // save
-            if let Some(saveout) = &self.saveout {
-                self.save(&img_rgba, saveout);
+            let saveout = self.saveout()?.join(format!("{}.png", string_now("-")));
+            match img_rgba.save(&saveout) {
+                Err(err) => tracing::error!("{} Saving failed: {:?}", CROSS_MARK, err),
+                Ok(_) => {
+                    tracing::info!("{} Annotated image saved to: {:?}", CHECK_MARK, saveout);
                 }
             }
+
+            vs.push(img_rgba);
+        }
+        Ok(vs)
     }
 
     /// Plot bounding bboxes and labels
     pub fn plot_bboxes(&self, img: &mut RgbaImage, bboxes: &[Bbox]) {
-        // bbox
         for bbox in bboxes.iter() {
+            // bbox
             let short_side_threshold =
                 bbox.width().min(bbox.height()) * self.bboxes_thickness_threshold;
             let thickness = self.bboxes_thickness.min(short_side_threshold as usize);
@@ -723,11 +751,11 @@ impl Annotator {
     /// Load custom font
     fn load_font(path: Option<&str>) -> Result<FontVec> {
         let path_font = match path {
-            None => auto_load("Arial.ttf", Some("fonts"))?,
+            None => Hub::new()?.fetch("fonts/Arial.ttf")?.commit()?,
             Some(p) => p.into(),
         };
         let buffer = std::fs::read(path_font)?;
-        Ok(FontVec::try_from_vec(buffer.to_owned()).unwrap())
+        Ok(FontVec::try_from_vec(buffer.to_owned())?)
     }
 
     /// Pick color from pallette

src/core/dataloader.rs

@@ -1,91 +1,336 @@
-use crate::{CHECK_MARK, SAFE_CROSS_MARK};
-use anyhow::{anyhow, bail, Result};
+use anyhow::{anyhow, Result};
 use image::DynamicImage;
+use indicatif::{ProgressBar, ProgressStyle};
 use std::collections::VecDeque;
 use std::path::{Path, PathBuf};
-use walkdir::{DirEntry, WalkDir};
+use std::sync::mpsc;
+use video_rs::{
+    encode::{Encoder, Settings},
+    time::Time,
+    Decoder, Url,
+};
 
-/// Dataloader for load images
-#[derive(Debug, Clone)]
-pub struct DataLoader {
-    pub paths: VecDeque<PathBuf>,
-    pub recursive: bool,
-    pub batch: usize,
+use crate::{
+    build_progress_bar, string_now, Dir, Hub, Location, MediaType, CHECK_MARK, CROSS_MARK,
+};
+
+type TempReturnType = (Vec<DynamicImage>, Vec<PathBuf>);
+
+pub struct DataLoaderIterator {
+    receiver: mpsc::Receiver<TempReturnType>,
+    progress_bar: Option<ProgressBar>,
 }
 
-impl Iterator for DataLoader {
-    type Item = (Vec<DynamicImage>, Vec<PathBuf>);
+impl Iterator for DataLoaderIterator {
+    type Item = TempReturnType;
 
     fn next(&mut self) -> Option<Self::Item> {
-        if self.paths.is_empty() {
+        match &self.progress_bar {
+            None => self.receiver.recv().ok(),
+            Some(progress_bar) => match self.receiver.recv().ok() {
+                Some(item) => {
+                    progress_bar.inc(1);
+                    Some(item)
+                }
+                None => {
+                    progress_bar.set_prefix("    Iterated");
+                    progress_bar.set_style(
+                        indicatif::ProgressStyle::with_template(crate::PROGRESS_BAR_STYLE_GREEN)
+                            .unwrap(),
+                    );
+                    progress_bar.finish();
                     None
-        } else {
-            let mut yis: Vec<DynamicImage> = Vec::new();
-            let mut yps: Vec<PathBuf> = Vec::new();
-            loop {
-                let path = self.paths.pop_front().unwrap();
-                match Self::try_read(&path) {
-                    Err(err) => {
-                        println!("{SAFE_CROSS_MARK} {err}");
                 }
-                    Ok(x) => {
-                        yis.push(x);
-                        yps.push(path);
-                    }
-                }
-                if self.paths.is_empty() || yis.len() == self.batch {
-                    break;
-                }
-            }
-            Some((yis, yps))
+            },
         }
     }
 }
 
-impl Default for DataLoader {
-    fn default() -> Self {
-        Self {
-            batch: 1,
-            recursive: false,
-            paths: Default::default(),
+impl IntoIterator for DataLoader {
+    type Item = TempReturnType;
+    type IntoIter = DataLoaderIterator;
+
+    fn into_iter(self) -> Self::IntoIter {
+        let progress_bar = if self.with_pb {
+            build_progress_bar(
+                self.nf / self.batch_size as u64,
+                "   Iterating",
+                Some(&format!("{:?}", self.media_type)),
+                crate::PROGRESS_BAR_STYLE_CYAN,
+            )
+            .ok()
+        } else {
+            None
+        };
+
+        DataLoaderIterator {
+            receiver: self.receiver,
+            progress_bar,
         }
     }
 }
 
+/// A structure designed to load and manage image, video, or stream data.
+/// It handles local file paths, remote URLs, and live streams, supporting both batch processing
+/// and optional progress bar display. The structure also supports video decoding through
+/// `video_rs` for video and stream data.
+pub struct DataLoader {
+    /// Queue of paths for images.
+    paths: Option<VecDeque<PathBuf>>,
+
+    /// Media type of the source (image, video, stream, etc.).
+    media_type: MediaType,
+
+    /// Batch size for iteration, determining how many files are processed at once.
+    batch_size: usize,
+
+    /// Buffer size for the channel, used to manage the buffer between producer and consumer.
+    bound: usize,
+
+    /// Receiver for processed data.
+    receiver: mpsc::Receiver<TempReturnType>,
+
+    /// Video decoder for handling video or stream data.
+    decoder: Option<video_rs::decode::Decoder>,
+
+    /// Number of images or frames; `u64::MAX` is used for live streams (indicating no limit).
+    nf: u64,
+
+    /// Flag indicating whether to display a progress bar.
+    with_pb: bool,
+}
+
 impl DataLoader {
-    pub fn load<P: AsRef<Path>>(mut self, source: P) -> Result<Self> {
-        self.paths = match source.as_ref() {
-            s if s.is_file() => VecDeque::from([s.to_path_buf()]),
-            s if s.is_dir() => WalkDir::new(s)
-                .into_iter()
-                .filter_entry(|e| !Self::_is_hidden(e))
-                .filter_map(|entry| match entry {
-                    Err(_) => None,
-                    Ok(entry) => {
-                        if entry.file_type().is_dir() {
-                            return None;
+    pub fn new(source: &str) -> Result<Self> {
+        let span = tracing::span!(tracing::Level::INFO, "DataLoader-new");
+        let _guard = span.enter();
+
+        // Number of frames or stream
+        let mut nf = 0;
+
+        // paths & media_type
+        let source_path = Path::new(source);
+        let (paths, media_type) = match source_path.exists() {
+            false => {
+                // remote
+                nf = 1;
+                (
+                    Some(VecDeque::from([source_path.to_path_buf()])),
+                    MediaType::from_url(source),
+                )
             }
-                        if !self.recursive && entry.depth() > 1 {
-                            return None;
+            true => {
+                // local
+                if source_path.is_file() {
+                    nf = 1;
+                    (
+                        Some(VecDeque::from([source_path.to_path_buf()])),
+                        MediaType::from_path(source_path),
+                    )
+                } else if source_path.is_dir() {
+                    let paths_sorted = Self::load_from_folder(source_path)?;
+                    nf = paths_sorted.len() as _;
+                    (
+                        Some(VecDeque::from(paths_sorted)),
+                        MediaType::Image(Location::Local),
+                    )
+                } else {
+                    (None, MediaType::Unknown)
                 }
-                        Some(entry.path().to_path_buf())
             }
-                })
-                .collect::<VecDeque<_>>(),
-            // s if s.starts_with("rtsp://") || s.starts_with("rtmp://") || s.starts_with("http://")|| s.starts_with("https://") => todo!(),
-            s if !s.exists() => bail!("{s:?} Not Exists"),
-            _ => todo!(),
         };
-        println!("{CHECK_MARK} Found file x{}", self.paths.len());
+
+        if let MediaType::Unknown = media_type {
+            anyhow::bail!("Could not locate the source path: {:?}", source_path);
+        }
+
+        // video decoder
+        let decoder = match &media_type {
+            MediaType::Video(Location::Local) => Some(Decoder::new(source_path)?),
+            MediaType::Video(Location::Remote) | MediaType::Stream => {
+                let location: video_rs::location::Location = source.parse::<Url>()?.into();
+                Some(Decoder::new(location)?)
+            }
+            _ => None,
+        };
+
+        // video & stream frames
+        if let Some(decoder) = &decoder {
+            nf = match decoder.frames() {
+                Err(_) => u64::MAX,
+                Ok(0) => u64::MAX,
+                Ok(x) => x,
+            }
+        }
+
+        // summary
+        tracing::info!("{} Found {:?} x{}", CHECK_MARK, media_type, nf,);
+
+        Ok(DataLoader {
+            paths,
+            media_type,
+            bound: 50,
+            receiver: mpsc::sync_channel(1).1,
+            batch_size: 1,
+            decoder,
+            nf,
+            with_pb: true,
+        })
+    }
+
+    pub fn with_bound(mut self, x: usize) -> Self {
+        self.bound = x;
+        self
+    }
+
+    pub fn with_batch(mut self, x: usize) -> Self {
+        self.batch_size = x;
+        self
+    }
+
+    pub fn with_progress_bar(mut self, x: bool) -> Self {
+        self.with_pb = x;
+        self
+    }
+
+    pub fn build(mut self) -> Result<Self> {
+        let (sender, receiver) = mpsc::sync_channel::<TempReturnType>(self.bound);
+        self.receiver = receiver;
+        let batch_size = self.batch_size;
+        let data = self.paths.take().unwrap_or_default();
+        let media_type = self.media_type.clone();
+        let decoder = self.decoder.take();
+
+        // Spawn the producer thread
+        std::thread::spawn(move || {
+            DataLoader::producer_thread(sender, data, batch_size, media_type, decoder);
+        });
+
         Ok(self)
     }
 
+    fn producer_thread(
+        sender: mpsc::SyncSender<TempReturnType>,
+        mut data: VecDeque<PathBuf>,
+        batch_size: usize,
+        media_type: MediaType,
+        mut decoder: Option<video_rs::decode::Decoder>,
+    ) {
+        let span = tracing::span!(tracing::Level::INFO, "DataLoader-producer-thread");
+        let _guard = span.enter();
+        let mut yis: Vec<DynamicImage> = Vec::with_capacity(batch_size);
+        let mut yps: Vec<PathBuf> = Vec::with_capacity(batch_size);
+
+        match media_type {
+            MediaType::Image(_) => {
+                while let Some(path) = data.pop_front() {
+                    match Self::try_read(&path) {
+                        Err(err) => {
+                            tracing::warn!("{} {:?} | {:?}", CROSS_MARK, path, err);
+                            continue;
+                        }
+                        Ok(img) => {
+                            yis.push(img);
+                            yps.push(path);
+                        }
+                    }
+                    if yis.len() == batch_size
+                        && sender
+                            .send((std::mem::take(&mut yis), std::mem::take(&mut yps)))
+                            .is_err()
+                    {
+                        break;
+                    }
+                }
+            }
+            MediaType::Video(_) | MediaType::Stream => {
+                if let Some(decoder) = decoder.as_mut() {
+                    let (w, h) = decoder.size();
+                    let frames = decoder.decode_iter();
+
+                    for frame in frames {
+                        match frame {
+                            Ok((ts, frame)) => {
+                                let rgb8: image::ImageBuffer<image::Rgb<u8>, Vec<u8>> =
+                                    match image::ImageBuffer::from_raw(
+                                        w as _,
+                                        h as _,
+                                        frame.into_raw_vec_and_offset().0,
+                                    ) {
+                                        Some(x) => x,
+                                        None => continue,
+                                    };
+                                let img = image::DynamicImage::from(rgb8);
+                                yis.push(img);
+                                yps.push(ts.to_string().into());
+
+                                if yis.len() == batch_size
+                                    && sender
+                                        .send((std::mem::take(&mut yis), std::mem::take(&mut yps)))
+                                        .is_err()
+                                {
+                                    break;
+                                }
+                            }
+                            Err(_) => break,
+                        }
+                    }
+                }
+            }
+            _ => todo!(),
+        }
+
+        // Deal with remaining data
+        if !yis.is_empty() && sender.send((yis, yps)).is_err() {
+            tracing::info!("Receiver dropped, stopping production");
+        }
+    }
+
+    pub fn load_from_folder<P: AsRef<std::path::Path>>(path: P) -> Result<Vec<std::path::PathBuf>> {
+        let mut paths: Vec<PathBuf> = std::fs::read_dir(path)?
+            .filter_map(|entry| entry.ok())
+            .filter_map(|entry| {
+                let path = entry.path();
+                if path.is_file() {
+                    Some(path)
+                } else {
+                    None
+                }
+            })
+            .collect();
+
+        paths.sort_by(|a, b| {
+            let a_name = a.file_name().and_then(|s| s.to_str());
+            let b_name = b.file_name().and_then(|s| s.to_str());
+
+            match (a_name, b_name) {
+                (Some(a_str), Some(b_str)) => natord::compare(a_str, b_str),
+                _ => std::cmp::Ordering::Equal,
+            }
+        });
+
+        Ok(paths)
+    }
+
     pub fn try_read<P: AsRef<Path>>(path: P) -> Result<DynamicImage> {
-        let img = image::ImageReader::open(&path)
+        let mut path = path.as_ref().to_path_buf();
+
+        // try to fetch from hub or local cache
+        if !path.exists() {
+            let p = Hub::new()?.fetch(path.to_str().unwrap())?.commit()?;
+            path = PathBuf::from(&p);
+        }
+        let img = Self::read_into_rgb8(path)?;
+        Ok(DynamicImage::from(img))
+    }
+
+    fn read_into_rgb8<P: AsRef<Path>>(path: P) -> Result<image::RgbImage> {
+        let path = path.as_ref();
+        let img = image::ImageReader::open(path)
             .map_err(|err| {
                 anyhow!(
                     "Failed to open image at {:?}. Error: {:?}",
-                    path.as_ref(),
+                    path.display(),
                     err
                 )
             })?
@@ -93,7 +338,7 @@ impl DataLoader {
             .map_err(|err| {
                 anyhow!(
                     "Failed to make a format guess based on the content: {:?}. Error: {:?}",
-                    path.as_ref(),
+                    path.display(),
                     err
                 )
             })?
@@ -101,33 +346,64 @@ impl DataLoader {
             .map_err(|err| {
                 anyhow!(
                     "Failed to decode image at {:?}. Error: {:?}",
-                    path.as_ref(),
+                    path.display(),
                     err
                 )
             })?
             .into_rgb8();
-        Ok(DynamicImage::from(img))
+        Ok(img)
     }
 
-    pub fn with_batch(mut self, x: usize) -> Self {
-        self.batch = x;
-        self
+    /// Convert images into a video
+    pub fn is2v<P: AsRef<Path>>(source: P, subs: &[&str], fps: usize) -> Result<()> {
+        let paths = Self::load_from_folder(source.as_ref())?;
+        if paths.is_empty() {
+            anyhow::bail!("No images found.");
+        }
+        let mut encoder = None;
+        let mut position = Time::zero();
+        let saveout = Dir::Currnet
+            .raw_path_with_subs(subs)?
+            .join(format!("{}.mp4", string_now("-")));
+        let pb = build_progress_bar(paths.len() as u64, "  Converting", saveout.to_str(),"{prefix:.cyan.bold} {msg} |{bar}| ({percent_precise}%, {human_pos}/{human_len}, {per_sec})")?;
+
+        // loop
+        for path in paths {
+            pb.inc(1);
+            let img = Self::read_into_rgb8(path)?;
+            let (w, h) = img.dimensions();
+
+            // build encoder at the 1st time
+            if encoder.is_none() {
+                let settings = Settings::preset_h264_yuv420p(w as _, h as _, false);
+                encoder = Some(Encoder::new(saveout.clone(), settings)?);
             }
 
-    pub fn with_recursive(mut self, x: bool) -> Self {
-        self.recursive = x;
-        self
+            // write video
+            if let Some(encoder) = encoder.as_mut() {
+                let raw_data = img.into_raw();
+                let frame = ndarray::Array3::from_shape_vec((h as usize, w as usize, 3), raw_data)
+                    .expect("Failed to create ndarray from raw image data");
+
+                // encode and update
+                encoder.encode(&frame, position)?;
+                position = position.aligned_with(Time::from_nth_of_a_second(fps)).add();
+            }
         }
 
-    pub fn paths(&self) -> &VecDeque<PathBuf> {
-        &self.paths
+        match &mut encoder {
+            Some(vencoder) => vencoder.finish()?,
+            None => anyhow::bail!("Found no video encoder."),
         }
 
-    fn _is_hidden(entry: &DirEntry) -> bool {
-        entry
-            .file_name()
-            .to_str()
-            .map(|s| s.starts_with('.'))
-            .unwrap_or(false)
+        // update
+        pb.set_prefix("  Downloaded");
+        pb.set_prefix("   Converted");
+        pb.set_style(ProgressStyle::with_template(
+            "{prefix:.green.bold} {msg} in {elapsed}",
+        )?);
+        pb.finish();
+
+        Ok(())
     }
 }

src/core/dir.rs

@@ -0,0 +1,119 @@
+/// Represents various directories on the system, including Home, Cache, Config, and more.
+#[derive(Debug)]
+pub enum Dir {
+    Home,
+    Cache,
+    Config,
+    Currnet,
+    Document,
+    Data,
+    Download,
+    Desktop,
+    Audio,
+    Picture,
+}
+
+impl Dir {
+    /// Retrieves the base path for the specified directory type, optionally appending the `usls` subdirectory.
+    ///
+    /// # Arguments
+    /// * `raw` - If `true`, returns the base path without adding the `usls` subdirectory.
+    ///
+    /// # Returns
+    /// * `Result<PathBuf>` - The base path for the directory.
+    fn get_path(&self, raw: bool) -> anyhow::Result<std::path::PathBuf> {
+        let base_path = match self {
+            Dir::Home => dirs::home_dir(),
+            Dir::Cache => dirs::cache_dir(),
+            Dir::Config => dirs::config_dir(),
+            Dir::Currnet => std::env::current_dir().ok(),
+            _ => None,
+        };
+
+        let mut path = base_path.ok_or_else(|| {
+            anyhow::anyhow!("Unsupported operating system. Now supports Linux, MacOS, Windows.")
+        })?;
+
+        if !raw {
+            if let Dir::Home = self {
+                path.push(".usls");
+            } else {
+                path.push("usls");
+            }
+        }
+        Ok(path)
+    }
+
+    /// Returns the default path for the `usls` directory, creating it automatically if it does not exist.
+    ///
+    /// Examples:
+    /// `~/.cache/usls`, `~/.config/usls`, `~/.usls`.
+    ///
+    /// # Returns
+    /// * `Result<PathBuf>` - The default `usls` directory path.
+    pub fn path(&self) -> anyhow::Result<std::path::PathBuf> {
+        let d = self.get_path(false)?;
+        self.create_directory(&d)?;
+        Ok(d)
+    }
+
+    /// Returns the raw path for the directory without adding the `usls` subdirectory.
+    ///
+    /// Examples:
+    /// `~/.cache`, `~/.config`, `~`.
+    ///
+    /// # Returns
+    /// * `Result<PathBuf>` - The raw directory path.
+    pub fn raw_path(&self) -> anyhow::Result<std::path::PathBuf> {
+        self.get_path(true)
+    }
+
+    /// Constructs a path to the `usls` directory with the provided subdirectories, creating it automatically.
+    ///
+    /// Examples:
+    /// `~/.cache/usls/sub1/sub2/sub3`, `~/.config/usls/sub1/sub2`, `~/.usls/sub1/sub2`.
+    ///
+    /// # Arguments
+    /// * `subs` - A slice of strings representing subdirectories to append.
+    ///
+    /// # Returns
+    /// * `Result<PathBuf>` - The resulting directory path.
+    pub fn path_with_subs(&self, subs: &[&str]) -> anyhow::Result<std::path::PathBuf> {
+        let mut d = self.get_path(false)?;
+        self.append_subs(&mut d, subs)?;
+        Ok(d)
+    }
+
+    /// Constructs a path to a specified directory with the provided subdirectories, creating it automatically.
+    ///
+    /// Examples:
+    /// `~/.cache/sub1/sub2/sub3`, `~/.config/sub1/sub2`, `~/sub1/sub2`.
+    ///
+    /// # Arguments
+    /// * `subs` - A slice of strings representing subdirectories to append.
+    ///
+    /// # Returns
+    /// * `Result<PathBuf>` - The resulting directory path.
+    pub fn raw_path_with_subs(&self, subs: &[&str]) -> anyhow::Result<std::path::PathBuf> {
+        let mut d = self.get_path(true)?;
+        self.append_subs(&mut d, subs)?;
+        Ok(d)
+    }
+
+    /// Appends subdirectories to the given base path and creates the directories if they don't exist.
+    fn append_subs(&self, path: &mut std::path::PathBuf, subs: &[&str]) -> anyhow::Result<()> {
+        for sub in subs {
+            path.push(sub);
+        }
+        self.create_directory(path)?;
+        Ok(())
+    }
+
+    /// Creates the specified directory if it does not exist.
+    fn create_directory(&self, path: &std::path::PathBuf) -> anyhow::Result<()> {
+        if !path.exists() {
+            std::fs::create_dir_all(path)?;
+        }
+        Ok(())
+    }
+}

src/core/hub.rs

@@ -0,0 +1,432 @@
+use anyhow::{Context, Result};
+use indicatif::{ProgressBar, ProgressStyle};
+use serde::{Deserialize, Serialize};
+use std::io::{Read, Write};
+use std::path::{Path, PathBuf};
+
+use crate::Dir;
+
+/// Represents a downloadable asset in a release
+#[derive(Clone, Debug, Serialize, Deserialize)]
+pub struct Asset {
+    pub name: String,
+    pub browser_download_url: String,
+    pub size: u64,
+}
+
+/// Represents a GitHub release
+#[derive(Clone, Debug, Serialize, Deserialize)]
+pub struct Release {
+    pub tag_name: String,
+    pub assets: Vec<Asset>,
+}
+
+/// Manages interactions with a GitHub repository's releases
+pub struct Hub {
+    /// github api
+    _gh_api_release: String,
+
+    /// GitHub repository owner
+    owner: String,
+
+    /// GitHub repository name          
+    repo: String,
+
+    /// Optional list of releases fetched from GitHub
+    releases: Option<Vec<Release>>,
+
+    /// Path to cache file
+    cache: PathBuf,
+
+    /// Optional release tag to be used
+    tag: Option<String>,
+
+    /// Filename for the asset, used in cache management
+    file_name: Option<String>,
+    file_size: Option<u64>,
+
+    /// Full URL constructed for downloading the asset
+    url: Option<String>,
+
+    /// Local path where the asset will be stored
+    path: PathBuf,
+
+    /// Directory to store the downloaded file
+    to: Dir,
+
+    /// Download timeout in seconds
+    timeout: u64,
+
+    /// Time to live (cache duration)
+    ttl: std::time::Duration,
+
+    /// Maximum attempts for downloading
+    max_attempts: u32,
+}
+
+impl std::fmt::Debug for Hub {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        f.debug_struct("Hub")
+            .field("owner", &self.owner)
+            .field("repo", &self.repo)
+            .field("cache", &self.cache)
+            .field("path", &self.path)
+            .field("releases", &self.releases.as_ref().map(|x| x.len()))
+            .field("ttl", &self.ttl)
+            .field("max_attempts", &self.max_attempts)
+            .finish()
+    }
+}
+
+impl Default for Hub {
+    fn default() -> Self {
+        let owner = "jamjamjon".to_string();
+        let repo = "assets".to_string();
+        let _gh_api_release = format!("https://api.github.com/repos/{}/{}/releases", owner, repo);
+
+        Self {
+            owner,
+            repo,
+            _gh_api_release,
+            url: None,
+            path: PathBuf::new(),
+            to: Dir::Cache,
+            tag: None,
+            file_name: None,
+            file_size: None,
+            releases: None,
+            cache: PathBuf::new(),
+            timeout: 3000,
+            max_attempts: 3,
+            ttl: std::time::Duration::from_secs(10 * 60),
+        }
+    }
+}
+
+impl Hub {
+    pub fn new() -> Result<Self> {
+        let mut to = Dir::Cache;
+        let cache = to
+            .path()
+            .or_else(|_| {
+                to = Dir::Home;
+                to.path()
+            })?
+            .join("cache_releases");
+
+        Ok(Self {
+            to,
+            cache,
+            ..Default::default()
+        })
+    }
+
+    pub fn with_owner(mut self, owner: &str) -> Self {
+        self.owner = owner.to_string();
+        self
+    }
+
+    pub fn with_repo(mut self, repo: &str) -> Self {
+        self.repo = repo.to_string();
+        self
+    }
+
+    pub fn with_ttl(mut self, x: u64) -> Self {
+        self.ttl = std::time::Duration::from_secs(x);
+        self
+    }
+
+    pub fn with_timeout(mut self, x: u64) -> Self {
+        self.timeout = x;
+        self
+    }
+
+    pub fn with_max_attempts(mut self, x: u32) -> Self {
+        self.max_attempts = x;
+        self
+    }
+
+    pub fn fetch(mut self, s: &str) -> Result<Self> {
+        // try to fetch from hub or local cache
+        let p = PathBuf::from(s);
+        match p.exists() {
+            true => self.path = p,
+            false => {
+                // check local cache 1st
+                let p_cache = self.cache.with_file_name(s);
+                if p_cache.exists() {
+                    self.path = p_cache;
+                } else {
+                    // check remote list then
+                    match s.split_once('/') {
+                        Some((tag, file_name)) => {
+                            // Extract tag and file from input string
+                            self.tag = Some(tag.to_string());
+                            self.file_name = Some(file_name.to_string());
+
+                            // Check if releases are already loaded in memory
+                            if self.releases.is_none() {
+                                self.releases = Some(self.connect_remote()?);
+                            }
+
+                            if let Some(releases) = &self.releases {
+                                // Validate the tag
+                                let tags: Vec<&str> =
+                                    releases.iter().map(|x| x.tag_name.as_str()).collect();
+                                if !tags.contains(&tag) {
+                                    anyhow::bail!(
+                                        "Hub tag '{}' not found in releases. Available tags: {:?}",
+                                        tag,
+                                        tags
+                                    );
+                                }
+
+                                // Validate the file
+                                if let Some(release) = releases.iter().find(|r| r.tag_name == tag) {
+                                    let files: Vec<&str> =
+                                        release.assets.iter().map(|x| x.name.as_str()).collect();
+                                    if !files.contains(&file_name) {
+                                        anyhow::bail!(
+                                            "Hub file '{}' not found in tag '{}'. Available files: {:?}",
+                                            file_name,
+                                            tag,
+                                            files
+                                        );
+                                    } else {
+                                        for f_ in release.assets.iter() {
+                                            if f_.name.as_str() == file_name {
+                                                self.url = Some(f_.browser_download_url.clone());
+                                                self.file_size = Some(f_.size);
+
+                                                break;
+                                            }
+                                        }
+                                    }
+                                }
+                                self.path = self.to.path_with_subs(&[tag])?.join(file_name);
+                            }
+                        }
+                        _ => anyhow::bail!(
+                        "Download failed due to invalid format. Expected: <tag>/<file>, got: {}",
+                        s
+                    ),
+                    }
+                }
+            }
+        }
+
+        Ok(self)
+    }
+
+    /// Fetch releases from GitHub and cache them
+    fn fetch_and_cache_releases(url: &str, cache_path: &Path) -> Result<String> {
+        let response = ureq::get(url)
+            .set("User-Agent", "my-app")
+            .call()
+            .context("Failed to fetch releases from remote")?;
+
+        if response.status() != 200 {
+            anyhow::bail!(
+                "Failed to fetch releases from remote ({}): status {} - {}",
+                url,
+                response.status(),
+                response.status_text()
+            );
+        }
+
+        let body = response
+            .into_string()
+            .context("Failed to read response body")?;
+
+        // Ensure cache directory exists
+        let parent_dir = cache_path
+            .parent()
+            .context("Invalid cache path; no parent directory found")?;
+        std::fs::create_dir_all(parent_dir)
+            .with_context(|| format!("Failed to create cache directory: {:?}", parent_dir))?;
+
+        // Create temporary file
+        let mut temp_file = tempfile::NamedTempFile::new_in(parent_dir)
+            .context("Failed to create temporary cache file")?;
+
+        // Write data to temporary file
+        temp_file
+            .write_all(body.as_bytes())
+            .context("Failed to write to temporary cache file")?;
+
+        // Persist temporary file as the cache
+        temp_file.persist(cache_path).with_context(|| {
+            format!("Failed to persist temporary cache file to {:?}", cache_path)
+        })?;
+
+        Ok(body)
+    }
+
+    pub fn tags(&mut self) -> Option<Vec<&str>> {
+        if self.releases.is_none() {
+            self.releases = self.connect_remote().ok();
+        }
+
+        self.releases
+            .as_ref()
+            .map(|releases| releases.iter().map(|x| x.tag_name.as_str()).collect())
+    }
+
+    pub fn files(&mut self, tag: &str) -> Option<Vec<&str>> {
+        if self.releases.is_none() {
+            self.releases = self.connect_remote().ok();
+        }
+
+        self.releases.as_ref().map(|releases| {
+            releases
+                .iter()
+                .find(|r| r.tag_name == tag)
+                .map(|a| a.assets.iter().map(|x| x.name.as_str()).collect())
+        })?
+    }
+
+    pub fn connect_remote(&mut self) -> Result<Vec<Release>> {
+        let span = tracing::span!(tracing::Level::INFO, "OrtEngine-run");
+        let _guard = span.enter();
+
+        let should_download = if !self.cache.exists() {
+            tracing::info!("No cache found, fetching data from GitHub");
+            true
+        } else {
+            match std::fs::metadata(&self.cache)?.modified() {
+                Err(_) => {
+                    tracing::info!("Cannot get file modified time, fetching new data from GitHub");
+                    true
+                }
+                Ok(modified_time) => {
+                    if std::time::SystemTime::now().duration_since(modified_time)? < self.ttl {
+                        tracing::info!("Using cached data");
+                        false
+                    } else {
+                        tracing::info!("Cache expired, fetching new data from GitHub");
+                        true
+                    }
+                }
+            }
+        };
+
+        let body = if should_download {
+            Self::fetch_and_cache_releases(&self._gh_api_release, &self.cache)?
+        } else {
+            std::fs::read_to_string(&self.cache)?
+        };
+        let releases: Vec<Release> = serde_json::from_str(&body)?;
+        Ok(releases)
+    }
+
+    /// Commit the downloaded file, downloading if necessary
+    pub fn commit(&self) -> Result<String> {
+        if let Some(url) = &self.url {
+            // Download if the file does not exist or if the size of file does not match
+            if !self.path.is_file()
+                || self.path.is_file()
+                    && Some(std::fs::metadata(&self.path)?.len()) != self.file_size
+            {
+                let name = format!(
+                    "{}/{}",
+                    self.tag.as_ref().unwrap(),
+                    self.file_name.as_ref().unwrap()
+                );
+                Self::download(
+                    url.as_str(),
+                    &self.path,
+                    Some(&name),
+                    Some(self.timeout),
+                    Some(self.max_attempts),
+                )?;
+            }
+        }
+        self.path
+            .to_str()
+            .map(|s| s.to_string())
+            .with_context(|| format!("Failed to convert PathBuf: {:?} to String", self.path))
+    }
+
+    /// Download a file from a github release to a specified path with a progress bar
+    pub fn download<P: AsRef<Path> + std::fmt::Debug>(
+        src: &str,
+        dst: P,
+        prompt: Option<&str>,
+        timeout: Option<u64>,
+        max_attempts: Option<u32>,
+    ) -> Result<()> {
+        // TODO: other url, not just github release page
+
+        let max_attempts = max_attempts.unwrap_or(2);
+        let timeout_duration = std::time::Duration::from_secs(timeout.unwrap_or(2000));
+        let agent = ureq::AgentBuilder::new().try_proxy_from_env(true).build();
+
+        for i_try in 0..max_attempts {
+            let resp = agent
+                .get(src)
+                .timeout(timeout_duration)
+                .call()
+                .with_context(|| {
+                    format!(
+                        "Failed to download file from {}, timeout: {:?}",
+                        src, timeout_duration
+                    )
+                })?;
+            let ntotal = resp
+                .header("Content-Length")
+                .and_then(|s| s.parse::<u64>().ok())
+                .context("Content-Length header is missing or invalid")?;
+
+            let pb = ProgressBar::new(ntotal);
+            pb.set_style(
+                ProgressStyle::with_template(
+                    "{prefix:.cyan.bold} {msg} |{bar}| ({percent_precise}%, {binary_bytes}/{binary_total_bytes}, {binary_bytes_per_sec})",
+                )?
+                .progress_chars("██ "),
+            );
+            pb.set_prefix(if i_try == 0 {
+                "    Fetching"
+            } else {
+                " Re-Fetching"
+            });
+            pb.set_message(prompt.unwrap_or_default().to_string());
+
+            let mut reader = resp.into_reader();
+            let mut buffer = [0; 256];
+            let mut downloaded_bytes = 0usize;
+            let mut file = std::fs::File::create(&dst)
+                .with_context(|| format!("Failed to create destination file: {:?}", dst))?;
+
+            loop {
+                let bytes_read = reader.read(&mut buffer)?;
+                if bytes_read == 0 {
+                    break;
+                }
+                file.write_all(&buffer[..bytes_read])
+                    .context("Failed to write to file")?;
+                downloaded_bytes += bytes_read;
+                pb.inc(bytes_read as u64);
+            }
+
+            // check size
+            if downloaded_bytes as u64 != ntotal {
+                continue;
+            }
+
+            // update
+            pb.set_prefix("  Downloaded");
+            pb.set_style(ProgressStyle::with_template(
+                "{prefix:.green.bold} {msg} ({binary_total_bytes}) in {elapsed}",
+            )?);
+            pb.finish();
+
+            if i_try != max_attempts {
+                break;
+            } else {
+                anyhow::bail!("Exceeded the maximum number of download attempts");
+            }
+        }
+
+        Ok(())
+    }
+}

src/core/media.rs

@@ -0,0 +1,64 @@
+use crate::{AUDIO_EXTENSIONS, IMAGE_EXTENSIONS, STREAM_PROTOCOLS, VIDEO_EXTENSIONS};
+
+#[derive(Debug, Clone)]
+pub enum MediaType {
+    Image(Location),
+    Video(Location),
+    Audio(Location),
+    Stream,
+    Unknown,
+}
+
+#[derive(Debug, Clone)]
+pub enum Location {
+    Local,
+    Remote,
+}
+
+#[derive(Debug, Clone)]
+pub enum StreamType {
+    Pre,
+    Live,
+}
+
+impl MediaType {
+    pub fn from_path<P: AsRef<std::path::Path>>(path: P) -> Self {
+        let extension = path
+            .as_ref()
+            .extension()
+            .and_then(|ext| ext.to_str())
+            .unwrap_or("")
+            .to_lowercase();
+
+        if IMAGE_EXTENSIONS.contains(&extension.as_str()) {
+            MediaType::Image(Location::Local)
+        } else if VIDEO_EXTENSIONS.contains(&extension.as_str()) {
+            MediaType::Video(Location::Local)
+        } else if AUDIO_EXTENSIONS.contains(&extension.as_str()) {
+            MediaType::Audio(Location::Local)
+        } else {
+            MediaType::Unknown
+        }
+    }
+
+    pub fn from_url(url: &str) -> Self {
+        if IMAGE_EXTENSIONS
+            .iter()
+            .any(|&ext| url.ends_with(&format!(".{}", ext)))
+        {
+            MediaType::Image(Location::Remote)
+        } else if VIDEO_EXTENSIONS
+            .iter()
+            .any(|&ext| url.ends_with(&format!(".{}", ext)))
+        {
+            MediaType::Video(Location::Remote)
+        } else if STREAM_PROTOCOLS
+            .iter()
+            .any(|&protocol| url.starts_with(protocol))
+        {
+            MediaType::Stream
+        } else {
+            MediaType::Unknown
+        }
+    }
+}

src/core/mod.rs

@@ -1,8 +1,11 @@
 mod annotator;
 mod dataloader;
 mod device;
+mod dir;
 mod dynconf;
+mod hub;
 mod logits_sampler;
+mod media;
 mod metric;
 mod min_opt_max;
 pub mod onnx;
@@ -19,8 +22,11 @@ mod xs;
 pub use annotator::Annotator;
 pub use dataloader::DataLoader;
 pub use device::Device;
+pub use dir::Dir;
 pub use dynconf::DynConf;
+pub use hub::Hub;
 pub use logits_sampler::LogitsSampler;
+pub use media::*;
 pub use metric::Metric;
 pub use min_opt_max::MinOptMax;
 pub use ops::Ops;

src/core/options.rs

@@ -3,9 +3,8 @@
 use anyhow::Result;
 
 use crate::{
-    auto_load,
     models::{SamKind, SapiensTask, YOLOPreds, YOLOTask, YOLOVersion},
-    Device, MinOptMax,
+    Device, Hub, MinOptMax,
 };
 
 /// Options for building models
@@ -15,14 +14,14 @@ pub struct Options {
     pub device: Device,
     pub profile: bool,
     pub num_dry_run: usize,
-    pub i00: Option<MinOptMax>, // 1st input, axis 0, batch usually
-    pub i01: Option<MinOptMax>, // 1st input, axis 1
+    pub i00: Option<MinOptMax>, // the 1st input, axis 0, batch usually
+    pub i01: Option<MinOptMax>, // the 1st input, axis 1
     pub i02: Option<MinOptMax>,
     pub i03: Option<MinOptMax>,
     pub i04: Option<MinOptMax>,
     pub i05: Option<MinOptMax>,
-    pub i10: Option<MinOptMax>, // 2nd input, axis 0
-    pub i11: Option<MinOptMax>, // 2nd input, axis 1
+    pub i10: Option<MinOptMax>, // the 2nd input, axis 0
+    pub i11: Option<MinOptMax>, // the 2nd input, axis 1
     pub i12: Option<MinOptMax>,
     pub i13: Option<MinOptMax>,
     pub i14: Option<MinOptMax>,
@@ -101,7 +100,7 @@ impl Default for Options {
             onnx_path: String::new(),
             device: Device::Cuda(0),
             profile: false,
-            num_dry_run: 5,
+            num_dry_run: 3,
             i00: None,
             i01: None,
             i02: None,
@@ -182,8 +181,12 @@ impl Default for Options {
 }
 
 impl Options {
+    pub fn new() -> Self {
+        Default::default()
+    }
+
     pub fn with_model(mut self, onnx_path: &str) -> Result<Self> {
-        self.onnx_path = auto_load(onnx_path, Some("models"))?;
+        self.onnx_path = Hub::new()?.fetch(onnx_path)?.commit()?;
         Ok(self)
     }
 
@@ -268,7 +271,7 @@ impl Options {
     }
 
     pub fn with_vocab(mut self, vocab: &str) -> Result<Self> {
-        self.vocab = Some(auto_load(vocab, Some("tokenizers"))?);
+        self.vocab = Some(Hub::new()?.fetch(vocab)?.commit()?);
         Ok(self)
     }
 
@@ -278,7 +281,7 @@ impl Options {
     }
 
     pub fn with_tokenizer(mut self, tokenizer: &str) -> Result<Self> {
-        self.tokenizer = Some(auto_load(tokenizer, Some("tokenizers"))?);
+        self.tokenizer = Some(Hub::new()?.fetch(tokenizer)?.commit()?);
         Ok(self)
     }
 

src/core/ort_engine.rs

@@ -8,7 +8,8 @@ use prost::Message;
 use std::collections::HashSet;
 
 use crate::{
-    home_dir, human_bytes, onnx, Device, MinOptMax, Ops, Options, Ts, Xs, CHECK_MARK, CROSS_MARK, X,
+    build_progress_bar, human_bytes, onnx, Device, Dir, MinOptMax, Ops, Options, Ts, Xs,
+    CHECK_MARK, CROSS_MARK, X,
 };
 
 /// Ort Tensor Attrs: name, data_type, dims
@@ -37,6 +38,9 @@ pub struct OrtEngine {
 
 impl OrtEngine {
     pub fn new(config: &Options) -> Result<Self> {
+        let span = tracing::span!(tracing::Level::INFO, "OrtEngine-new");
+        let _guard = span.enter();
+
         // onnx graph
         let model_proto = Self::load_onnx(&config.onnx_path)?;
         let graph = match &model_proto.graph {
@@ -150,13 +154,13 @@ impl OrtEngine {
             }
             Device::Cuda(device_id) => {
                 Self::build_cuda(&builder, device_id).unwrap_or_else(|err| {
+                    tracing::warn!("{err}, Using cpu");
                     device = Device::Cpu(0);
-                    println!("{err}");
                 })
             }
             Device::CoreML(_) => Self::build_coreml(&builder).unwrap_or_else(|err| {
+                tracing::warn!("{err}, Using cpu");
                 device = Device::Cpu(0);
-                println!("{err}");
             }),
             Device::Cpu(_) => {
                 Self::build_cpu(&builder)?;
@@ -169,7 +173,7 @@ impl OrtEngine {
             .commit_from_file(&config.onnx_path)?;
 
         // summary
-        println!(
+        tracing::info!(
             "{CHECK_MARK} Backend: ONNXRuntime | Opset: {} | Device: {:?} | Params: {}",
             model_proto.opset_import[0].version,
             device,
@@ -200,6 +204,9 @@ impl OrtEngine {
         fp16_enable: bool,
         engine_cache_enable: bool,
     ) -> Result<()> {
+        let span = tracing::span!(tracing::Level::INFO, "OrtEngine-new");
+        let _guard = span.enter();
+
         // auto generate shapes
         let mut spec_min = String::new();
         let mut spec_opt = String::new();
@@ -228,22 +235,19 @@ impl OrtEngine {
             spec_opt += &s_opt;
             spec_max += &s_max;
         }
+        let p = Dir::Cache.path_with_subs(&["trt-cache"])?;
         let trt = TensorRTExecutionProvider::default()
             .with_device_id(device_id as i32)
             .with_int8(int8_enable)
             .with_fp16(fp16_enable)
             .with_engine_cache(engine_cache_enable)
-            .with_engine_cache_path(format!(
-                "{}/{}",
-                home_dir(None).to_str().unwrap(),
-                "trt-cache"
-            ))
+            .with_engine_cache_path(p.to_str().unwrap())
             .with_timing_cache(false)
             .with_profile_min_shapes(spec_min)
             .with_profile_opt_shapes(spec_opt)
             .with_profile_max_shapes(spec_max);
         if trt.is_available()? && trt.register(builder).is_ok() {
-            println!("\n🐢 Initial model serialization with TensorRT may require a wait...\n");
+            tracing::info!("🐢 Initial model serialization with TensorRT may require a wait...\n");
             Ok(())
         } else {
             anyhow::bail!("{CROSS_MARK} TensorRT initialization failed")
@@ -279,6 +283,15 @@ impl OrtEngine {
 
     pub fn dry_run(&mut self) -> Result<()> {
         if self.num_dry_run > 0 {
+            // pb
+            let pb = build_progress_bar(
+                self.num_dry_run as u64,
+                "      DryRun",
+                Some(&format!("{:?}", self.device)),
+                crate::PROGRESS_BAR_STYLE_CYAN,
+            )?;
+
+            // dummy inputs
             let mut xs = Vec::new();
             for i in self.inputs_minoptmax.iter() {
                 let mut x: Vec<usize> = Vec::new();
@@ -289,16 +302,27 @@ impl OrtEngine {
                 xs.push(X::from(x));
             }
             let xs = Xs::from(xs);
+
+            // run
             for _ in 0..self.num_dry_run {
+                pb.inc(1);
                 self.run(xs.clone())?;
             }
             self.ts.clear();
-            println!("{CHECK_MARK} Dryrun x{}", self.num_dry_run);
+
+            // update
+            pb.set_style(indicatif::ProgressStyle::with_template(
+                crate::PROGRESS_BAR_STYLE_GREEN,
+            )?);
+            pb.finish();
         }
         Ok(())
     }
 
     pub fn run(&mut self, xs: Xs) -> Result<Xs> {
+        let span = tracing::span!(tracing::Level::INFO, "OrtEngine-run");
+        let _guard = span.enter();
+
         // inputs dtype alignment
         let mut xs_ = Vec::new();
         let t_pre = std::time::Instant::now();
@@ -369,7 +393,7 @@ impl OrtEngine {
         if self.profile {
             let len = 10usize;
             let n = 4usize;
-            println!(
+            tracing::info!(
                 "[Profile] {:>len$.n$?} ({:>len$.n$?} avg) [alignment: {:>len$.n$?} ({:>len$.n$?} avg) | inference: {:>len$.n$?} ({:>len$.n$?} avg) | to_f32: {:>len$.n$?} ({:>len$.n$?} avg)]",
                 t_pre + t_run + t_post,
                 self.ts.avg(),
@@ -624,10 +648,7 @@ impl OrtEngine {
     pub fn try_fetch(&self, key: &str) -> Option<String> {
         match self.session.metadata() {
             Err(_) => None,
-            Ok(metadata) => match metadata.custom(key) {
-                Err(_) => None,
-                Ok(value) => value,
-            },
+            Ok(metadata) => metadata.custom(key).unwrap_or_default(),
         }
     }
 

src/core/vision.rs

@@ -25,6 +25,9 @@ pub trait Vision: Sized {
 
     /// Executes the full pipeline.
     fn forward(&mut self, xs: &[Self::Input], profile: bool) -> anyhow::Result<Vec<Y>> {
+        let span = tracing::span!(tracing::Level::INFO, "DataLoader-new");
+        let _guard = span.enter();
+
         let t_pre = std::time::Instant::now();
         let ys = self.preprocess(xs)?;
         let t_pre = t_pre.elapsed();
@@ -38,7 +41,9 @@ pub trait Vision: Sized {
         let t_post = t_post.elapsed();
 
         if profile {
-            println!("> Preprocess: {t_pre:?} | Execution: {t_exe:?} | Postprocess: {t_post:?}");
+            tracing::info!(
+                "> Preprocess: {t_pre:?} | Execution: {t_exe:?} | Postprocess: {t_post:?}"
+            );
         }
 
         Ok(ys)

src/lib.rs

@@ -1,42 +1,71 @@
-//! A Rust library integrated with ONNXRuntime, providing a collection of **Computer Vision** and **Vision-Language** models.
+//! **usls** is a Rust library integrated with **ONNXRuntime** that provides a collection of state-of-the-art models for **Computer Vision** and **Vision-Language** tasks, including:
 //!
-//! # Supported Models
-//!
-//! - [YOLOv5](https://github.com/ultralytics/yolov5): Object Detection, Instance Segmentation, Classification
-//! - [YOLOv6](https://github.com/meituan/YOLOv6): Object Detection
-//! - [YOLOv7](https://github.com/WongKinYiu/yolov7): Object Detection
-//! - [YOLOv8](https://github.com/ultralytics/ultralytics): Object Detection, Instance Segmentation, Classification, Oriented Object Detection, Keypoint Detection
-//! - [YOLOv9](https://github.com/WongKinYiu/yolov9): Object Detection
-//! - [YOLOv10](https://github.com/THU-MIG/yolov10): Object Detection
-//! - [RT-DETR](https://arxiv.org/abs/2304.08069): Object Detection
-//! - [FastSAM](https://github.com/CASIA-IVA-Lab/FastSAM): Instance Segmentation
-//! - [SAM](https://github.com/facebookresearch/segment-anything): Segmentation Anything
-//! - [MobileSAM](https://github.com/ChaoningZhang/MobileSAM): Segmentation Anything
-//! - [EdgeSAM](https://github.com/chongzhou96/EdgeSAM): Segmentation Anything
-//! - [SAM-HQ](https://github.com/SysCV/sam-hq): Segmentation Anything
-//! - [YOLO-World](https://github.com/AILab-CVC/YOLO-World): Object Detection
-//! - [DINOv2](https://github.com/facebookresearch/dinov2): Vision-Self-Supervised
-//! - [CLIP](https://github.com/openai/CLIP): Vision-Language
-//! - [BLIP](https://github.com/salesforce/BLIP): Vision-Language
-//! - [DB](https://arxiv.org/abs/1911.08947): Text Detection
-//! - [SVTR](https://arxiv.org/abs/2205.00159): Text Recognition
-//! - [RTMO](https://github.com/open-mmlab/mmpose/tree/main/projects/rtmo): Keypoint Detection
-//! - [YOLOPv2](https://arxiv.org/abs/2208.11434): Panoptic Driving Perception
-//! - [Depth-Anything (v1, v2)](https://github.com/LiheYoung/Depth-Anything): Monocular Depth Estimation
-//! - [MODNet](https://github.com/ZHKKKe/MODNet): Image Matting
-//! - [Sapiens](https://arxiv.org/abs/2408.12569): Human-centric Vision Tasks
+//! - **YOLO Models**: [YOLOv5](https://github.com/ultralytics/yolov5), [YOLOv6](https://github.com/meituan/YOLOv6), [YOLOv7](https://github.com/WongKinYiu/yolov7), [YOLOv8](https://github.com/ultralytics/ultralytics), [YOLOv9](https://github.com/WongKinYiu/yolov9), [YOLOv10](https://github.com/THU-MIG/yolov10)
+//! - **SAM Models**: [SAM](https://github.com/facebookresearch/segment-anything), [SAM2](https://github.com/facebookresearch/segment-anything-2), [MobileSAM](https://github.com/ChaoningZhang/MobileSAM), [EdgeSAM](https://github.com/chongzhou96/EdgeSAM), [SAM-HQ](https://github.com/SysCV/sam-hq), [FastSAM](https://github.com/CASIA-IVA-Lab/FastSAM)
+//! - **Vision Models**: [RTDETR](https://arxiv.org/abs/2304.08069), [RTMO](https://github.com/open-mmlab/mmpose/tree/main/projects/rtmo), [DB](https://arxiv.org/abs/1911.08947), [SVTR](https://arxiv.org/abs/2205.00159), [Depth-Anything-v1-v2](https://github.com/LiheYoung/Depth-Anything), [DINOv2](https://github.com/facebookresearch/dinov2), [MODNet](https://github.com/ZHKKKe/MODNet), [Sapiens](https://arxiv.org/abs/2408.12569)
+//! - **Vision-Language Models**: [CLIP](https://github.com/openai/CLIP), [BLIP](https://arxiv.org/abs/2201.12086), [GroundingDINO](https://github.com/IDEA-Research/GroundingDINO), [YOLO-World](https://github.com/AILab-CVC/YOLO-World)
 //!
 //! # Examples
 //!
-//! [All Demos Here](https://github.com/jamjamjon/usls/tree/main/examples)
+//! Refer to [All Runnable Demos](https://github.com/jamjamjon/usls/tree/main/examples)
 //!
-//! # Using Provided Models for Inference
+//! # Quick Start
 //!
-//! #### 1. Build Model
-//! Using provided [`models`] with [`Options`]
+//! The following demo shows how to build a `YOLO` with [`Options`], load `image(s)`, `video` and `stream` with [`DataLoader`], and annotate the model's inference results with [`Annotator`].
 //!
-//! ```rust, no_run
-//! use usls::{ models::YOLO, Annotator, DataLoader, Options, Vision, COCO_CLASS_NAMES_80};
+//! ```ignore
+//! use usls::{models::YOLO, Annotator, DataLoader, Options, Vision, YOLOTask, YOLOVersion};
+//!
+//! fn main() -> anyhow::Result<()> {
+//!     // Build model with Options
+//!     let options = Options::new()
+//!         .with_trt(0)
+//!         .with_model("yolo/v8-m-dyn.onnx")?
+//!         .with_yolo_version(YOLOVersion::V8) // YOLOVersion: V5, V6, V7, V8, V9, V10, RTDETR
+//!         .with_yolo_task(YOLOTask::Detect) // YOLOTask: Classify, Detect, Pose, Segment, Obb
+//!         .with_i00((1, 1, 4).into())
+//!         .with_i02((0, 640, 640).into())
+//!         .with_i03((0, 640, 640).into())
+//!         .with_confs(&[0.2]);
+//!     let mut model = YOLO::new(options)?;
+//!
+//!     // Build DataLoader to load image(s), video, stream
+//!     let dl = DataLoader::new(
+//!         "./assets/bus.jpg", // local image
+//!         // "images/bus.jpg",  // remote image
+//!         // "../set-negs",  // local images (from folder)
+//!         // "../hall.mp4",  // local video
+//!         // "http://commondatastorage.googleapis.com/gtv-videos-bucket/sample/BigBuckBunny.mp4",  // remote video
+//!         // "rtsp://admin:kkasd1234@192.168.2.217:554/h264/ch1/",  // stream
+//!     )?
+//!     .with_batch(3)  // iterate with batch_size = 3
+//!     .build()?;
+//!
+//!     // Build annotator
+//!     let annotator = Annotator::new().with_saveout("YOLO-Demo");
+//!
+//!     // Run and Annotate images
+//!     for (xs, _) in dl {
+//!         let ys = model.forward(&xs, false)?;
+//!         annotator.annotate(&xs, &ys);
+//!     }
+//!
+//!     Ok(())
+//! }
+//! ```
+//!
+
+//! # What's More
+//!
+//! This guide covers the process of using provided models for inference, including how to build a model, load data, annotate results, and retrieve the outputs. Click the sections below to expand for detailed instructions.
+//!
+//! <details>
+//! <summary>Build the Model</summary>
+//!
+//! To build a model, you can use the provided [models] with [Options]:
+//!
+//! ```ignore
+//! use usls::{models::YOLO, Annotator, DataLoader, Options, Vision};
 //!
 //! let options = Options::default()
 //!     .with_yolo_version(YOLOVersion::V8)  // YOLOVersion: V5, V6, V7, V8, V9, V10, RTDETR
@@ -45,107 +74,149 @@
 //! let mut model = YOLO::new(options)?;
 //! ```
 //!
-//! - Use `CUDA`, `TensorRT`, or `CoreML`
+//! **And there're many options provided by [Options]**
 //!
-//! ```rust, no_run
+//! - **Choose Execution Provider:**  
+//!   Select `CUDA` (default), `TensorRT`, or `CoreML`:
+//!
+//! ```ignore
 //! let options = Options::default()
-//!     .with_cuda(0) // using CUDA by default
+//!     .with_cuda(0)
 //!     // .with_trt(0)
 //!     // .with_coreml(0)
 //!     // .with_cpu();
 //! ```
 //!
-//! - Dynamic Input Shapes
+//! - **Dynamic Input Shapes:**  
+//!   Specify dynamic shapes with [MinOptMax]:
 //!
-//! ```rust, no_run
+//! ```ignore
 //! let options = Options::default()
-//!     .with_i00((1, 2, 4).into()) // dynamic batch
-//!     .with_i02((416, 640, 800).into()) // dynamic height
-//!     .with_i03((416, 640, 800).into()); // dynamic width
+//!     .with_i00((1, 2, 4).into()) // batch(min=1, opt=2, max=4)
+//!     .with_i02((416, 640, 800).into()) // height(min=416, opt=640, max=800)
+//!     .with_i03((416, 640, 800).into()); // width(min=416, opt=640, max=800)
 //! ```
 //!
-//! - Set Confidence Thresholds for Each Category
+//! - **Set Confidence Thresholds:**  
+//!   Adjust thresholds for each category:
 //!
-//! ```rust, no_run
+//! ```ignore
 //! let options = Options::default()
 //!     .with_confs(&[0.4, 0.15]); // class_0: 0.4, others: 0.15
 //! ```
 //!
-//! - Set Class Names
+//! - **Set Class Names:**  
+//!   Provide class names if needed:
 //!
-//! ```rust, no_run
+//! ```ignore
 //! let options = Options::default()
 //!     .with_names(&COCO_CLASS_NAMES_80);
 //! ```
 //!
-//! More options can be found in the [`Options`] documentation.
+//! **More options are detailed in the [Options] documentation.**  
 //!   
-//! #### 2. Load Images
 //!
-//! Ensure that the input image is RGB type.
+//! </details>
 //!
-//! - Using [`image::ImageReader`] or [`DataLoader`] to Load One Image
+//! <details>
+//! <summary>Load Images, Video and Stream</summary>
 //!
-//! ```rust, no_run
-//! let x = vec![DataLoader::try_read("./assets/bus.jpg")?];
-//! // or
+//! - **Load a Single Image**  
+//! Use [DataLoader::try_read] to load an image from a local file or remote source:
+//!
+//! ```ignore
+//! let x = DataLoader::try_read("./assets/bus.jpg")?; // from local
+//! let x = DataLoader::try_read("images/bus.jpg")?; // from remote
+//! ```
+//!
+//! Alternatively, use [image::ImageReader] directly:
+//!
+//! ```ignore
 //! let x = image::ImageReader::open("myimage.png")?.decode()?;
 //! ```
 //!
-//! - Using [`DataLoader`] to Load a Batch of Images
+//! - **Load Multiple Images, Videos, or Streams**  
+//! Create a [DataLoader] instance for batch processing:
 //!
-//! ```rust, no_run
-//! let dl = DataLoader::default()
-//!     .with_batch(4)
-//!     .load("./assets")?;
+//! ```ignore
+//! let dl = DataLoader::new(
+//!     "./assets/bus.jpg", // local image
+//!     // "images/bus.jpg",  // remote image
+//!     // "../set-negs",  // local images (from folder)
+//!     // "../hall.mp4",  // local video
+//!     // "http://commondatastorage.googleapis.com/gtv-videos-bucket/sample/BigBuckBunny.mp4",  // remote video
+//!     // "rtsp://admin:kkasd1234@192.168.2.217:554/h264/ch1/",  // stream
+//! )?
+//! .with_batch(3)  // iterate with batch_size = 3
+//! .build()?;
+//!
+//! // Iterate through the data
+//! for (xs, _) in dl {}
 //! ```
 //!
-//! #### 3. (Optional) Annotate Results with [`Annotator`]
+//! - **Convert Images to Video**  
+//! Use [DataLoader::is2v] to create a video from a sequence of images:
 //!
-//! ```rust, no_run
+//! ```ignore
+//! let fps = 24;
+//! let image_folder = "runs/YOLO-DataLoader";
+//! let saveout = ["runs", "is2v"];
+//! DataLoader::is2v(image_folder, &saveout, fps)?;
+//! ```
+//!
+//! </details>
+//!
+//! <details>
+//! <summary>Annotate Inference Results</summary>
+//!
+//! - **Create an Annotator Instance**
+//!
+//! ```ignore
 //! let annotator = Annotator::default();
 //! ```
 //!
-//! - Set Saveout Name
+//! - **Set Saveout Name:**
 //!
-//! ```rust, no_run
+//! ```ignore
 //! let annotator = Annotator::default()
 //!     .with_saveout("YOLOs");
 //! ```
 //!
-//! - Set Bboxes Line Width
+//! - **Set Bounding Box Line Width:**
 //!
-//! ```rust, no_run
+//! ```ignore
 //! let annotator = Annotator::default()
 //!     .with_bboxes_thickness(4);
 //! ```
 //!
-//! - Disable Mask Plotting
+//! - **Disable Mask Plotting**
 //!
-//! ```rust, no_run
+//! ```ignore
 //! let annotator = Annotator::default()
 //!     .without_masks(true);
 //! ```
 //!
-//! More options can be found in the [`Annotator`] documentation.
+//! - **Perform Inference and nnotate the results**  
 //!
-//!     
-//! #### 4. Run and Annotate
-//!
-//! ```rust, no_run
+//! ```ignore
 //! for (xs, _paths) in dl {
 //!     let ys = model.run(&xs)?;
 //!     annotator.annotate(&xs, &ys);
 //! }
 //! ```
 //!
-//! #### 5. Get Results
+//! More options are detailed in the [Annotator] documentation.
 //!
-//! The inference outputs of provided models will be saved to a [`Vec<Y>`].
+//! </details>
 //!
-//! - For Example, Get Detection Bboxes with `y.bboxes()`
+//! <details>
+//! <summary>Retrieve Model's Inference Results</summary>
 //!
-//! ```rust, no_run
+//! Retrieve the inference outputs, which are saved in a [`Vec<Y>`]:
+//!
+//! - **Get Detection Bounding Boxes**
+//!
+//! ```ignore
 //! let ys = model.run(&xs)?;
 //! for y in ys {
 //!     // bboxes
@@ -165,11 +236,16 @@
 //! }
 //! ```
 //!
-//! # Also, You Can Implement Your Own Model with [`OrtEngine`] and [`Options`]
+//! </details>
 //!
-//! [`OrtEngine`] provides ONNX model loading, metadata parsing, dry_run, inference, and other functions, supporting EPs such as CUDA, TensorRT, CoreML, etc. You can use it as the ONNXRuntime engine for building models.
+//! <details>
+//! <summary>Custom Model Implementation</summary>
 //!
-//! Refer to [Demo: Depth-Anything](https://github.com/jamjamjon/usls/blob/main/src/models/depth_anything.rs) for more details.
+//! You can also implement your own model using [OrtEngine] and [Options]. [OrtEngine] supports ONNX model loading, metadata parsing, dry_run, inference, and other functions, with execution providers such as CUDA, TensorRT, CoreML, etc.
+//!
+//! For more details, refer to the [Demo: Depth-Anything](https://github.com/jamjamjon/usls/blob/main/src/models/depth_anything.rs).
+//!
+//! </details>
 
 mod core;
 pub mod models;

src/models/blip.rs

@@ -5,8 +5,7 @@ use std::io::Write;
 use tokenizers::Tokenizer;
 
 use crate::{
-    auto_load, Embedding, LogitsSampler, MinOptMax, Ops, Options, OrtEngine, TokenizerStream, Xs,
-    X, Y,
+    Embedding, LogitsSampler, MinOptMax, Ops, Options, OrtEngine, TokenizerStream, Xs, X, Y,
 };
 
 #[derive(Debug)]
@@ -31,13 +30,9 @@ impl Blip {
             visual.width().to_owned(),
         );
 
-        let tokenizer = match options_textual.tokenizer {
-            Some(x) => x,
-            None => match auto_load("tokenizer-blip.json", Some("tokenizers")) {
-                Err(err) => anyhow::bail!("No tokenizer's file found: {:?}", err),
-                Ok(x) => x,
-            },
-        };
+        let tokenizer = options_textual
+            .tokenizer
+            .ok_or(anyhow::anyhow!("No tokenizer file found"))?;
         let tokenizer = match Tokenizer::from_file(tokenizer) {
             Err(err) => anyhow::bail!("Failed to build tokenizer: {:?}", err),
             Ok(x) => x,

src/models/clip.rs

@@ -3,7 +3,7 @@ use image::DynamicImage;
 use ndarray::Array2;
 use tokenizers::{PaddingDirection, PaddingParams, PaddingStrategy, Tokenizer};
 
-use crate::{auto_load, Embedding, MinOptMax, Ops, Options, OrtEngine, Xs, X, Y};
+use crate::{Embedding, MinOptMax, Ops, Options, OrtEngine, Xs, X, Y};
 
 #[derive(Debug)]
 pub struct Clip {
@@ -29,13 +29,10 @@ impl Clip {
             visual.inputs_minoptmax()[0][3].to_owned(),
         );
 
-        let tokenizer = match options_textual.tokenizer {
-            Some(x) => x,
-            None => match auto_load("tokenizer-clip.json", Some("tokenizers")) {
-                Err(err) => anyhow::bail!("No tokenizer's file found: {:?}", err),
-                Ok(x) => x,
-            },
-        };
+        let tokenizer = options_textual
+            .tokenizer
+            .ok_or(anyhow::anyhow!("No tokenizer file found"))?;
+
         let mut tokenizer = match Tokenizer::from_file(tokenizer) {
             Err(err) => anyhow::bail!("Failed to build tokenizer: {:?}", err),
             Ok(x) => x,

src/models/dinov2.rs

@@ -28,8 +28,8 @@ impl Dinov2 {
             engine.inputs_minoptmax()[0][3].to_owned(),
         );
         let which = match options.onnx_path {
-            s if s.contains("b14") => Model::B,
-            s if s.contains("s14") => Model::S,
+            s if s.contains('b') => Model::B,
+            s if s.contains('s') => Model::S,
             _ => todo!(),
         };
         let hidden_size = match which {

src/models/grounding_dino.rs

@@ -1,4 +1,4 @@
-use crate::{auto_load, Bbox, DynConf, MinOptMax, Ops, Options, OrtEngine, Xs, X, Y};
+use crate::{Bbox, DynConf, MinOptMax, Ops, Options, OrtEngine, Xs, X, Y};
 use anyhow::Result;
 use image::DynamicImage;
 use ndarray::{s, Array, Axis};
@@ -27,13 +27,9 @@ impl GroundingDINO {
         );
         let context_length = options.context_length.unwrap_or(256);
         // let special_tokens = ["[CLS]", "[SEP]", ".", "?"];
-        let tokenizer = match options.tokenizer {
-            Some(x) => x,
-            None => match auto_load("tokenizer-groundingdino.json", Some("tokenizers")) {
-                Err(err) => anyhow::bail!("No tokenizer's file found: {:?}", err),
-                Ok(x) => x,
-            },
-        };
+        let tokenizer = options
+            .tokenizer
+            .ok_or(anyhow::anyhow!("No tokenizer file found"))?;
         let tokenizer = match Tokenizer::from_file(tokenizer) {
             Err(err) => anyhow::bail!("Failed to build tokenizer: {:?}", err),
             Ok(x) => x,

src/models/yolo.rs

@@ -32,6 +32,9 @@ impl Vision for YOLO {
     type Input = DynamicImage;
 
     fn new(options: Options) -> Result<Self> {
+        let span = tracing::span!(tracing::Level::INFO, "YOLO-new");
+        let _guard = span.enter();
+
         let mut engine = OrtEngine::new(&options)?;
         let (batch, height, width) = (
             engine.batch().to_owned(),
@@ -49,7 +52,7 @@ impl Vision for YOLO {
                 "segment" => Some(YOLOTask::Segment),
                 "obb" => Some(YOLOTask::Obb),
                 s => {
-                    println!("YOLO Task: {s:?} is unsupported");
+                    tracing::error!("YOLO Task: {s:?} is unsupported");
                     None
                 }
             }));
@@ -135,7 +138,7 @@ impl Vision for YOLO {
         let iou = options.iou.unwrap_or(0.45);
 
         // Summary
-        println!("YOLO Task: {:?}, Version: {:?}", task, version);
+        tracing::info!("YOLO Task: {:?}, Version: {:?}", task, version);
 
         engine.dry_run()?;
 

src/utils/mod.rs

@@ -1,8 +1,7 @@
-use anyhow::{anyhow, Result};
+#![allow(dead_code)]
+
 use indicatif::{ProgressBar, ProgressStyle};
 use rand::{distributions::Alphanumeric, thread_rng, Rng};
-use std::io::{Read, Write};
-use std::path::{Path, PathBuf};
 
 pub mod colormap256;
 pub mod names;
@@ -10,11 +9,25 @@ pub mod names;
 pub use colormap256::*;
 pub use names::*;
 
-pub(crate) const GITHUB_ASSETS: &str =
-    "https://github.com/jamjamjon/assets/releases/download/v0.0.1";
 pub(crate) const CHECK_MARK: &str = "✅";
 pub(crate) const CROSS_MARK: &str = "❌";
 pub(crate) const SAFE_CROSS_MARK: &str = "❎";
+pub(crate) const NETWORK_PREFIXES: &[&str] = &[
+    "http://", "https://", "ftp://", "ftps://", "sftp://", "rtsp://", "mms://", "mmsh://",
+    "rtmp://", "rtmps://", "file://",
+];
+pub(crate) const IMAGE_EXTENSIONS: &[&str] = &["jpg", "jpeg", "png", "gif", "bmp", "tiff", "webp"];
+pub(crate) const VIDEO_EXTENSIONS: &[&str] = &[
+    "mp4", "avi", "mkv", "mov", "wmv", "flv", "webm", "mpeg", "mpg", "m4v", "m4p",
+];
+pub(crate) const AUDIO_EXTENSIONS: &[&str] = &["mp3", "wav", "flac", "aac", "ogg", "wma"];
+pub(crate) const STREAM_PROTOCOLS: &[&str] = &[
+    "rtsp://", "rtsps://", "rtspu://", "rtmp://", "rtmps://", "hls://", "http://", "https://",
+];
+pub(crate) const PROGRESS_BAR_STYLE_CYAN: &str =
+    "{prefix:.cyan.bold} {msg} {human_pos}/{human_len} |{bar}| {elapsed_precise}";
+pub(crate) const PROGRESS_BAR_STYLE_GREEN: &str =
+    "{prefix:.green.bold} {msg} {human_pos}/{human_len} |{bar}| {elapsed_precise}";
 
 pub fn human_bytes(size: f64) -> String {
     let units = ["B", "KB", "MB", "GB", "TB", "PB", "EB"];
@@ -30,71 +43,6 @@ pub fn human_bytes(size: f64) -> String {
     format!("{:.1} {}", size, units[unit_index])
 }
 
-pub(crate) fn auto_load<P: AsRef<Path>>(src: P, sub: Option<&str>) -> Result<String> {
-    let src = src.as_ref();
-    let p = if src.is_file() {
-        src.into()
-    } else {
-        let sth = src.file_name().unwrap().to_str().unwrap();
-        let mut p = home_dir(sub);
-        p.push(sth);
-        if !p.is_file() {
-            download(
-                &format!("{}/{}", GITHUB_ASSETS, sth),
-                &p,
-                Some(sth.to_string().as_str()),
-            )?;
-        }
-        p
-    };
-    Ok(p.to_str().unwrap().to_string())
-}
-
-/// `download` sth from src to dst
-pub fn download<P: AsRef<Path> + std::fmt::Debug>(
-    src: &str,
-    dst: P,
-    prompt: Option<&str>,
-) -> Result<()> {
-    let resp = ureq::AgentBuilder::new()
-        .try_proxy_from_env(true)
-        .build()
-        .get(src)
-        .timeout(std::time::Duration::from_secs(2000))
-        .call()
-        .map_err(|err| anyhow!("Failed to download. {err:?}"))?;
-    let ntotal = resp
-        .header("Content-Length")
-        .and_then(|s| s.parse::<u64>().ok())
-        .expect("Content-Length header should be present on archive response");
-    let pb = ProgressBar::new(ntotal);
-    pb.set_style(
-            ProgressStyle::with_template(
-                "{prefix:.bold} {msg:.dim} [{bar:.blue.bright/white.dim}] {binary_bytes}/{binary_total_bytes} ({binary_bytes_per_sec}, {percent_precise}%, {elapsed})"
-            )
-            .unwrap()
-            .progress_chars("#>-"));
-    pb.set_prefix(String::from("\n🐢 Downloading"));
-    pb.set_message(prompt.unwrap_or_default().to_string());
-    let mut reader = resp.into_reader();
-    let mut buffer = [0; 256];
-    let mut downloaded_bytes = 0usize;
-    let mut f = std::fs::File::create(&dst).expect("Failed to create file");
-    loop {
-        let bytes_read = reader.read(&mut buffer)?;
-        if bytes_read == 0 {
-            break;
-        }
-        pb.inc(bytes_read as u64);
-        f.write_all(&buffer[..bytes_read])?;
-        downloaded_bytes += bytes_read;
-    }
-    assert_eq!(downloaded_bytes as u64, ntotal);
-    pb.finish();
-    println!();
-    Ok(())
-}
-
 pub(crate) fn string_random(n: usize) -> String {
     thread_rng()
         .sample_iter(&Alphanumeric)
@@ -112,33 +60,16 @@ pub(crate) fn string_now(delimiter: &str) -> String {
     t_now.format(&fmt).to_string()
 }
 
-#[allow(dead_code)]
-pub(crate) fn config_dir() -> PathBuf {
-    match dirs::config_dir() {
-        Some(mut d) => {
-            d.push("usls");
-            if !d.exists() {
-                std::fs::create_dir_all(&d).expect("Failed to create usls config directory.");
-            }
-            d
-        }
-        None => panic!("Unsupported operating system. Now support Linux, MacOS, Windows."),
-    }
-}
+pub fn build_progress_bar(
+    n: u64,
+    prefix: &str,
+    msg: Option<&str>,
+    style_temp: &str,
+) -> anyhow::Result<ProgressBar> {
+    let pb = ProgressBar::new(n);
+    pb.set_style(ProgressStyle::with_template(style_temp)?.progress_chars("██ "));
+    pb.set_prefix(prefix.to_string());
+    pb.set_message(msg.unwrap_or_default().to_string());
 
-#[allow(dead_code)]
-pub(crate) fn home_dir(sub: Option<&str>) -> PathBuf {
-    match dirs::home_dir() {
-        Some(mut d) => {
-            d.push(".usls");
-            if let Some(sub) = sub {
-                d.push(sub);
-            }
-            if !d.exists() {
-                std::fs::create_dir_all(&d).expect("Failed to create usls home directory.");
-            }
-            d
-        }
-        None => panic!("Unsupported operating system. Now support Linux, MacOS, Windows."),
-    }
+    Ok(pb)
 }

src/ys/y.rs

@@ -95,13 +95,6 @@ impl Y {
     /// # Returns
     ///
     /// * `Self` - The updated struct instance with the new probabilities set.
-    ///
-    /// # Examples
-    ///
-    /// ```
-    /// let probs = Prob::default();
-    /// let y = Y::default().with_probs(&probs);
-    /// ```
     pub fn with_probs(mut self, probs: &Prob) -> Self {
         self.probs = Some(probs.clone());
         self