Besides the skeleton annotations, you also need RGB videos to train RGBPoseConv3D. You need to download them from the official website of NTURGB+D
除了骨架注释,您还需要 RGB 视频来训练 RGBPoseConv3D。您需要从 NTURGB+D 的官方网站下载它们。
and put these videos in
$PYSKL/data/nturgbd_raw
并将这些视频放在 $PYSKL/data/nturgbd_raw
After that, you need to use the provided script to compress the raw videos (from 1920x1080 to 960x540) and switch the suffix to .mp4:
之后,您需要使用提供的脚本压缩原始视频(从 1920x1080 到 960x540)并将后缀切换为 .mp4:
# That step is mandatory, unless you know how to modify the code & config to make it work for raw videos!
python compress_nturgbd.py
After that, you will find processed videos in $PYSKL/data/nturgbd_videos, named like S001C001P001R001A001.mp4.
之后,您将在 $PYSKL/data/nturgbd_videos 中找到经过处理的视频,名称类似于 S001C001P001R001A001.mp4。
参考:
pyskl/configs/rgbpose_conv3d/compress_nturgbd.py at main · kennymckormick/pyskl · GitHub