Simple implementation of YOLOv7 pre-training model deployment based on OpenVINO toolkit

●一、YOLOv7简介●

官方版的YOLOv7相同体量下比YOLOv5精度更高,速度快120%（FPS）,比 YOLOX 快180%（FPS）,比 Dual-Swin-T 快1200%（FPS）,比 ConvNext 快550%（FPS）,比 SWIN-L快500%（FPS）.在5FPS到160FPS的范围内,无论是速度或是精度,YOLOv7都超过了目前已知的检测器,并且在GPU V100上进行测试, 精度为56.8% AP的模型可达到30 FPS（batch=1）以上的检测速率,与此同时,这是目前唯一一款在如此高精度下仍能超过30FPS的检测器.

论文链接：https://arxiv.org/abs/2207.02696

代码链接：https://github.com/WongKinYiu/yolov7

●二、预训练模型准备●

• 模型权重下载

  可以从官方githubThe link provided in the repository is based on the downloadCOCO数据集的YOLOv7预训练模型权重.

Model	Test Size	APtest	AP50test	AP75test	batch 1 fps	batch 32 average time
YOLOv7	640	51.4%	69.7%	55.9%	161 fps	2.8 ms
YOLOv7-x	640	53.1%	71.2%	57.8%	114 fps	4.3ms
YOLOv7-W6	1280	54.9%	72.6%	60.1%	84 fps	7.6 ms
YOLOv7-E6	1280	56.0%	73.5%	61.2%	56 fps	12.3 ms
YOLOv7-D6	1280	56.6%	74.0%	61.8%	44 fps	15.0 ms
YOLOv7-E6E	1280	56.8%	74.4%	62.1%	36 fps	18.7 ms

• 模型转换

  可以从官方githubThe link provided in the repository is based on the downloadCOCO数据集的YOLOv7预训练模型权重.

# 下载YOLOv7官方仓库：

$ git clone [email protected]:WongKinYiu/yolov7.git

$ cd yolov7/models

$ python export.py --weights yolov7.pt

●三、模型部署●

通过Netron工具打开yolov7.onnx文件后可以看到,The official pre-trained modeloutputThe section contains the prediction results of the three feature layers,Therefore, it needs to be based on the prior box of each layer（anchor）After adjusting the output data,再进行堆叠.

由于YOLOv7The model pre- and post-processing basic sumsYOLOv5一致,Most data processing modules can be reused directly.话不多说直接上代码：

●六、运行结果●

运行python示例后,会在本地dataGenerate code in the directorybounding box以及label的图片,Here we use the horse data attached to the official repository for testing,具体结果如下：

# 运行代码

$ python YOLOV7.py -i horse.jpg -m yolov7.onnx

Github地址：

https://github.com/OpenVINO-dev-contest/YOLOv7_OpenVINO

--END--