README.md

metadata

license: apache-2.0
tags:
  - RyzenAI
  - object-detection
  - vision
  - YOLO
  - Pytorch
datasets:
  - COCO
metrics:
  - mAP

YOLOv5s model trained on COCO

YOLOv5s is the small version of YOLOv5 model trained on COCO object detection (118k annotated images) at resolution 640x640. It was released in https://github.com/ultralytics/yolov5.

We develop a modified version that could be supported by AMD Ryzen AI.

Model description

YOLOv5 🚀 is the world's most loved vision AI, representing Ultralytics open-source research into future vision AI methods, incorporating lessons learned and best practices evolved over thousands of hours of research and development.

Intended uses & limitations

You can use the raw model for object detection. See the model hub to look for all available YOLOv5 models.

How to use

Installation

Follow Ryzen AI Installation to prepare the environment for Ryzen AI. Run the following script to install pre-requisites for this model.

pip install -r requirements.txt 

Data Preparation (optional: for accuracy evaluation)

The dataset MSCOCO2017 contains 118287 images for training and 5000 images for validation.

Download COCO dataset and create directories in your code like this:

└── datasets
     └── coco
           ├── annotations
           |   ├── instances_val2017.json
           |   └── ...
           ├── labels
           |   ├── val2017
           |   |   ├── 000000000139.txt
           |       ├── 000000000285.txt
           |       └── ...
           ├── images
           |   ├── val2017
           |   |   ├── 000000000139.jpg
           |       ├── 000000000285.jpg
           └── val2017.txt

1. put the val2017 image folder under images directory or use a softlink
2. the labels folder and val2017.txt above are generate by general_json2yolo.py
3. modify the coco.yaml like this:

path: /path/to/your/datasets/coco  # dataset root dir
train: train2017.txt  # train images (relative to 'path') 118287 images
val: val2017.txt  # val images (relative to 'path') 5000 images

Test & Evaluation

• Code snippet from onnx_inference.py on how to use

    args = make_parser().parse_args()
    onnx_path = args.model
    onnx_model = onnxruntime.InferenceSession(onnx_path)
    grid = np.load("./grid.npy", allow_pickle=True)
    anchor_grid = np.load("./anchor_grid.npy", allow_pickle=True)
    path = args.image_path 
    new_path = args.output_path
    conf_thres, iou_thres, classes, agnostic_nms, max_det = 0.25, 0.45, None, False, 1000

    img0 = cv2.imread(path)
    img = pre_process(img0)
    onnx_input = {onnx_model.get_inputs()[0].name: img}
    onnx_output = onnx_model.run(None, onnx_input)
    onnx_output = post_process(onnx_output)
    pred = non_max_suppression(
        onnx_output[0], conf_thres, iou_thres, classes, agnostic_nms, max_det=max_det
    )
    colors = Colors()
    det = pred[0]
    im0 = img0.copy()
    annotator = Annotator(im0, line_width=2, example=str(names))
    if len(det):
        # Rescale boxes from img_size to im0 size
        det[:, :4] = scale_coords(img.shape[2:], det[:, :4], im0.shape).round()

        # Write results
        for *xyxy, conf, cls in reversed(det):
            c = int(cls)  # integer class
            label = f"{names[c]} {conf:.2f}"
            annotator.box_label(xyxy, label, color=colors(c, True))
    # Stream results
    im0 = annotator.result()
    cv2.imwrite(new_path, im0)

• Run inference for a single image

python onnx_inference.py -m ./yolov5s_qat.onnx -i /Path/To/Your/Image --ipu --provider_config /Path/To/Your/Provider_config

Note: vaip_config.json is located at the setup package of Ryzen AI (refer to Installation)

• Test accuracy of the quantized model

python onnx_eval.py -m ./yolov5s_qat.onnx --ipu --provider_config /Path/To/Your/Provider_config

Performance

Metric	Accuracy on IPU
[email protected]:0.95	0.356

@software{glenn_jocher_2021_5563715,
  author       = {Glenn Jocher et. al.},
  title        = {{ultralytics/yolov5: v6.0 - YOLOv5n 'Nano' models, 
                   Roboflow integration, TensorFlow export, OpenCV
                   DNN support}},
  month        = oct,
  year         = 2021,
  publisher    = {Zenodo},
  version      = {v6.0},
  doi          = {10.5281/zenodo.5563715},
  url          = {https://doi.org/10.5281/zenodo.5563715}
}