撰文｜FengWen、BBuf

1

模型导出

这个教程用来解释如何导出一个训练好的OneFlow YOLOv5模型到 ONNX。欢迎大家到这里查看本篇文章的完整版本：start.oneflow.org/oneflow-yol…

2

开始之前

克隆工程并在 Python>3.7.0 的环境中安装 requiresments.txt , OneFlow 请选择 nightly 版本或者 >0.9 版本 。模型和数据可以从源码中自动下载。

git clone https://github.com/Oneflow-Inc/one-yolov5.git
cd one-yolov5
pip install -r requirements.txt  # install

3

格式

YOLOv5支持多种模型格式的导出，并基于特定模型对应的框架获得推理加速。

4

导出训练好的 YOLOv5 模型

下面的命令把预训练的 YOLOV5s 模型导出为 ONNX 格式。yolov5s 是小模型，是可用的模型里面第二小的。其它选项是 yolov5n ，yolov5m，yolov5l，yolov5x ，以及他们的 P6 对应项比如 yolov5s6 ，或者你自定义的模型，即 runs/exp/weights/best 。有关可用模型的更多信息，可以参考我们的README

python export.py --weights ../yolov5s/ --include onnx

提示: 添加 –half 以 FP16 半精度导出模型以实现更小的文件大小。

输出：

export: data=data/coco128.yaml, weights=['../yolov5s/'], imgsz=[640, 640], batch_size=1, device=cpu, half=False, inplace=False, train=False, keras=False, optimize=False, int8=False, dynamic=False, simplify=False, opset=12, verbose=False, workspace=4, nms=False, agnostic_nms=False, topk_per_class=100, topk_all=100, iou_thres=0.45, conf_thres=0.25, include=['onnx']
YOLOv5  270ac92 Python-3.8.11 oneflow-0.8.1+cu117.git.0c70a3f6be CPU

Fusing layers... 
YOLOv5s summary: 157 layers, 7225885 parameters, 229245 gradients

OneFlow: starting from ../yolov5s with output shape (1, 25200, 85) (112.9 MB)

ONNX: starting export with onnx 1.12.0...
Converting model to onnx....
Using opset <onnx, 12>
Optimizing ONNX model
After optimization: Const +17 (73->90), Identity -1 (1->0), Unsqueeze -60 (60->0), output -1 (1->0), variable -60 (127->67)
Succeed converting model, save model to ../yolov5s.onnx
<class 'tuple'>
Comparing result between oneflow and onnx....
Compare succeed!
ONNX: export success, saved as ../yolov5s.onnx (28.0 MB)

Export complete (24.02s)
Results saved to /home/zhangxiaoyu
Detect:          python detect.py --weights ../yolov5s.onnx 
Validate:        python val.py --weights ../yolov5s.onnx 
OneFlow Hub:     model = flow.hub.load('OneFlow-Inc/one-yolov5', 'custom', '../yolov5s.onnx')
Visualize:       https://netron.app

导出的 onnx 模型使用 Netron Viewer 进行可视化的结果如下：

5

导出模型的示例用法

detect.py 可以对导出的模型进行推理：

python path/to/detect.py --weights yolov5s/                  # OneFlow
                                   yolov5s.onnx               # ONNX Runtime or OpenCV DNN with --dnn
                                   yolov5s.xml                # OpenVINO
                                   yolov5s.engine             # TensorRT
                                   yolov5s.mlmodel            # CoreML (macOS only)
                                   yolov5s_saved_model        # TensorFlow SavedModel
                                   yolov5s.pb                 # TensorFlow GraphDef
                                   yolov5s.tflite             # TensorFlow Lite
                                   yolov5s_edgetpu.tflite     # TensorFlow Edge TPU

val.py 可以对导出的模型进行验证：

python path/to/val.py --weights    yolov5s/                  # OneFlow
                                   yolov5s.onnx               # ONNX Runtime or OpenCV DNN with --dnn
                                   yolov5s.xml                # OpenVINO
                                   yolov5s.engine             # TensorRT
                                   yolov5s.mlmodel            # CoreML (macOS only)
                                   yolov5s_saved_model        # TensorFlow SavedModel
                                   yolov5s.pb                 # TensorFlow GraphDef
                                   yolov5s.tflite             # TensorFlow Lite
                                   yolov5s_edgetpu.tflite     # TensorFlow Edge TPU

6

ONNX Runtime 推理

基于 onnx 模型使用 onnxruntime 进行推理：

python3 detect.py --weights ../yolov5s/yolov5s.onnx

输出：

detect: weights=['../yolov5s/yolov5s.onnx'], source=data/images, data=data/coco128.yaml, imgsz=[640, 640], conf_thres=0.25, iou_thres=0.45, max_det=1000, device=, view_img=False, save_txt=False, save_conf=False, save_crop=False, nosave=False, classes=None, agnostic_nms=False, augment=False, visualize=False, update=False, project=runs/detect, name=exp, exist_ok=False, line_thickness=3, hide_labels=False, hide_conf=False, half=False, dnn=False
YOLOv5  270ac92 Python-3.8.11 oneflow-0.8.1+cu117.git.0c70a3f6be 
Loading ../yolov5s/yolov5s.onnx for ONNX Runtime inference...
detect.py:159: DeprecationWarning: In future, it will be an error for 'np.bool_' scalars to be interpreted as an index
  s += f"{n} {names[int(c)]}{'s' * (n > 1)}, "  # add to string
image 1/2 /home/zhangxiaoyu/one-yolov5/data/images/bus.jpg: 640x640 4 persons, 1 bus, Done. (0.009s)
image 2/2 /home/zhangxiaoyu/one-yolov5/data/images/zidane.jpg: 640x640 2 persons, 2 ties, Done. (0.011s)
0.5ms pre-process, 10.4ms inference, 4.8ms NMS per image at shape (1, 3, 640, 640)
Results saved to runs/detect/exp14

参考文章

github.com/ultralytics…

欢迎 Star、试用 OneFlow 最新版本：

github.com/Oneflow-Inc…