仓库用于做yolo系列的QAT训练；

model	map@50-95	map@50
yolov5s.pt	0.374	0.572
yolov5s_8w8f_qdq.onnx	0.367	0.567
compiled.axmodel	0.368	0.567

环境安装

pip install -r requirements.txt

我们发现 onnxruntime 和 onnxscript 的其他版本可能引起精度误差和导出错误，因此pytorch==2.6; onnxruntime==1.21.0 onnxscript==0.4.0 是必须的。

数据集路径修改

修改 coco.yaml 中的数据集路径;

训练

python qat_base_ptq.py --data coco.yaml --epochs 50 --weights yolov5s.pt --cfg yolov5s.yaml --batch-size 96 --device 3 --lsq --hyp hyp.no-augmentation.yaml --save-period 10

测试

qdq 模型测试: python val.py --weights yolov5s_8w8f_qdq.onnx --data coco.yaml --training-onnx

Class     Images  Instances          P          R      mAP50   mAP50-95: 100%|██████████| 5000/5000 [02:00<00:00, 41.47it/s]
  all       5000      36335       0.67      0.518      0.562      0.365

 Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.367
 Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.567
 Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.396
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.211
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.414
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.485
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.308
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.507
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.558
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.371
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.617
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.712

浮点模型测试：python val.py --weights yolov5s.pt --data coco.yaml

Class     Images  Instances          P          R      mAP50   mAP50-95: 100%|██████████| 157/157 [00:35<00:00,  4.43it/s]
  all       5000      36335      0.672      0.519      0.566      0.371

 Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.374
 Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.572
 Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.402
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.211
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.423
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.490
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.311
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.516
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.566
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.378
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.625
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.722

编译

在 compile目录下提供了 compiled.axmodel.onnx config.json val_on_board.py 三个文件；

编译命令: pulsar2 build --input yolov5s_8w8f_qdq.onnx --config config.json --output_dir output

在支持ubuntu的板子上，可以直接上板模型测试 : python val_on_board.py --model compiled.axmodel --dataset ../dataset/coco/

Parameters:
  --model: compiled.axmodel
  --dataset_path: datasets/coco/
  --save_results_path: cache/results.json
  --conf: 0.001
  --iou: 0.6
  --num_class: 80

[INFO] Available providers:  ['AxEngineExecutionProvider']
[INFO] Using provider: AxEngineExecutionProvider
[INFO] Chip type: ChipType.MC50
[INFO] VNPU type: VNPUType.DISABLED
[INFO] Engine version: 2.12.0s
[INFO] Model type: 0 (single core)
[INFO] Compiler version: 4.2-dirty 9c83050e-dirty
sample contains 5000 data

......

 Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.368
 Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.567
 Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.397
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.215
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.414
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.483
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.308
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.508
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.558
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.382
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.616
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.712