【pose】EfficientPose 测试过程记录

下载工程代码

参考教程 https://github.com/daniegr/EfficientPose

git clone https://github.com/daniegr/EfficientPose.git

注意事项

To achieve the optimal precision provided by the software, please adhere to the following three principles:

1. Ensure there is only one person present in the image/video （单人）
2. Ensure that the full body of the person is clearly visible, and near the centre of the image/video frame（全身清晰可见）
3. Avoid that the subject is occluded by other objects, even partial occlusion is discouraged（没有遮挡）

环境配置

Assuming you have Python (>= 3.6) and FFMPEG (>= 4.2) preinstalled(https://blog.csdn.net/u014470361/article/details/90705336)

command

• conda create -n EPOSE python==3.7
• source activate EPOSE
• cd EfficientPose
• pip install -r requirements.txt -i https://pypi.tuna.tsinghua.edu.cn/simple
  • 提前修改 opencv-python==3.4.2.17

• track.py中 修改测试video的路径和 framework_name
  file_path = "/home/cody/PycharmProjects/yolov5/video/result_1.mp4"
  model_name = 'I' #可以选择 "I","II","III","IV","RT",
  framework_name = 'keras'

运行track.py

Image processed in 0.338 seconds ,这个耗时忍不了，不会考虑在此基础上优化，精度一般，不抗造

<按理说此处应该放图>