Haodong Duan
Haodong Duan
Hi, [Gbouna](https://github.com/Gbouna), The problem is that PoseConv3D only applies to 2D skeletons, so you can not directly use 3D skeletons for training and testing (which will lead to the reported...
Hi, @Sunkey2333 , You can train / test existing algorithm supported by PYSKL with this framework, and you can also support your own with the framework (will be quite easy)...
Yeah, you can try the RTMPose (implemented in MMPose), I think it can detect human skeletons in real-time.
Currently, we have not supported GCN models + Kinetics, we will support them shortly.
Hi, Yu-0920, Are you using a linux computer? If not please try to install on a linux distribution.
It runs on Windows, but only the demo and testing part (training not supported). I can provide a conda environment for Windows later this day.
I'm not sure (perhaps not, since PYSKL is using a different format compared to MMAction2).
Hi, @icost77 , To ensemble to two streams, you need to first save the inference result (.pkl files that include the predicted probabilities for every action class and all videos)....
Yes I think, as long as the demo code can successfully call the API to inference the recognition results, which corresponds to this line: https://github.com/kennymckormick/pyskl/blob/e36e39c365b1ac038da0f558c0e6a3de2c328cea/demo/demo_skeleton.py#L287
Hi, [Gbouna](https://github.com/Gbouna), I just tried the latest version of the main branch on my computer, and it works fine. It looks like the error is due to that the `img_shape`...