usls/README.md

usls

A Rust library integrated with ONNXRuntime, providing a collection of Computer Vison and Vision-Language models including YOLOv8 (Classification, Segmentation, Detection and Pose Detection), YOLOv9, RTDETR, CLIP, DINOv2, FastSAM, YOLO-World, BLIP, PaddleOCR and others. Many execution providers are supported, sunch as CUDA, TensorRT and CoreML.

Supported Models

Model	Example	CUDA f32	CUDA f16	TensorRT f32	TensorRT f16
YOLOv8-detection	demo	✅	✅	✅	✅
YOLOv8-pose	demo	✅	✅	✅	✅
YOLOv8-classification	demo	✅	✅	✅	✅
YOLOv8-segmentation	demo	✅	✅	✅	✅
YOLOv8-OBB	TODO	TODO	TODO	TODO	TODO
YOLOv9	demo	✅	✅	✅	✅
RT-DETR	demo	✅	✅	✅	✅
FastSAM	demo	✅	✅	✅	✅
YOLO-World	demo	✅	✅	✅	✅
DINOv2	demo	✅	✅	✅	✅
CLIP	demo	✅	✅	✅ visual ❌ textual	✅ visual ❌ textual
BLIP	demo	✅	✅	✅ visual ❌ textual	✅ visual ❌ textual
DB(Text Detection)	demo	✅	❌	✅	✅
SVTR(Text Recognition)	demo	✅	❌	✅	✅

Solution Models

Additionally, this repo also provides some solution models such as pedestrian fall detection, head detection, trash detection, and more.

Model	Example
text detection (PPOCR-det v3, v4) 通用文本检测	demo
text recognition (PPOCR-rec v3, v4) 中英文-文本识别	demo
face-landmark detection 人脸 & 关键点检测	demo
head detection 人头检测	demo
fall detection 摔倒检测	demo
trash detection 垃圾检测	demo

Demo

cargo run -r --example yolov8   # fastsam, yolov9, blip, clip, dinov2, yolo-world...

Integrate into your own project

1. Install ort

check ort guide

For Linux or MacOS users

• Firstly, download from latest release from ONNXRuntime Releases
• Then linking

  export ORT_DYLIB_PATH=/Users/qweasd/Desktop/onnxruntime-osx-arm64-1.17.1/lib/libonnxruntime.1.17.1.dylib
  

2. Add usls as a dependency to your project's Cargo.toml

cargo add --git https://github.com/jamjamjon/usls

3. Set Options and build model

let options = Options::default()
    .with_model("../models/yolov8m-seg-dyn-f16.onnx");
let mut model = YOLO::new(&options)?;

• If you want to run your model with TensorRT or CoreML

  let options = Options::default()
      .with_trt(0) // using cuda by default
      // .with_coreml(0) 
  

• If your model has dynamic shapes

  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
  

• If you want to set a confidence level for each category

  let options = Options::default()
      .with_confs(&[0.4, 0.15]) // person: 0.4, others: 0.15
  

• Go check Options for more model options.

4. Prepare inputs, and then you're ready to go

• Build DataLoader to load images

let dl = DataLoader::default()
    .with_batch(model.batch.opt as usize)
    .load("./assets/")?;

for (xs, _paths) in dl {
    let _y = model.run(&xs)?;
}

• Or simply read one image

let x = vec![DataLoader::try_read("./assets/bus.jpg")?];
let y = model.run(&x)?;

5. Annotate and save results

let annotator = Annotator::default().with_saveout("YOLOv8");
annotator.annotate(&x, &y);

Script: converte ONNX model from float32 to float16

import onnx
from pathlib import Path
from onnxconverter_common import float16

model_f32 = "onnx_model.onnx"
model_f16 = float16.convert_float_to_float16(onnx.load(model_f32))
saveout = Path(model_f32).with_name(Path(model_f32).stem + "-f16.onnx")
onnx.save(model_f16, saveout)