Closed hunminkim98 closed 4 months ago
Awesome, thank you!
device='cpu'
in tests.py.install_requires
in setup.cfg:
rtmlib
onnxruntime-gpu
openvino
(we also need to remove opencv-python, as rtmlib installs it too and installing it twice sometimes leads to errors)
I'll switch to discord for further discussion :)
This is the same as a Discord post.
It now supports the following: Automatically download any selected models. mode = 'performance', 'balanced', 'lightweight' skeleton_type = 'body', 'body_and_foot', 'fullbody' install rtmlib as part of Pose2Sim's installation process. use rtmlib's functions in Pose2Sim's PoseEstimation.py. OpenPose style in all models except halpe26(alreay has neck keypoint). However, it is quite slow. Inference with video and images Save the result as jsons Save the result as video Realtime-visualization
I think after some final preparations for integration (cleanup of config.toml and other structural optimizations), raw video to opensim is now ready.
Additionally, rtmlib seems to support webcam, could this be useful for Pose2Sim?
I just merged it into the test branch and I'll test it now!
Hello, here is a request for PoseEstimator!
The name of the folder that includes raw video is set to 'pose_raw' I haven't tested with images yet, but videos should work. For now, I'll keep testing and improving it!
Thanks!