mediapipe

mediapipe copied to clipboard

I observed that it only detects the closest person to the cam in a video, but I want to detect a person who is some distance away and not the closest one? How do I do that?

btw the person I want to detect can be considered to be center of the video

Thank you

+1 I see the issue of detecting multiple persons is repeatedly raised from time to time and the solution suggested (modify the graph) isn't really practical as there is no any tutorial on how to make such graph modification, in addition to the requirement to recompile the whole model. I think MP can do a huge favor to the community by adding support for multi-pose detection! the approach proposed in https://github.com/google/mediapipe/issues/1691#issuecomment-1103730953 can be a good start.

BTW, other libraries such as OpenPose do support the multi-pose detection, it would be better if MP would add such a capability to save people from searching alternative libraries.

Hi @ArtfulMinx , Could you please check this https://github.com/google/mediapipe/issues/3539#issuecomment-1193091309 comments and confirm still if you have any error.

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you.

@sureshdagooglecom none of the comments there describes the steps needed to enable multiple persons detection, it says that the graph need to be changed. How to change it? is there any explanation or tutorial on how to build and use a modified graph?

Closing as stale. Please reopen if you'd like to work on this further.

Are you satisfied with the resolution of your issue? Yes No

Hi @ArtfulMinx , Pose landmarks model is intended to be used with an image of a single person for now. We can't guarantee the same quality if there will be another person in the same model input image. Hopefully, in the future, we'll extend the model's ability to work on this case. We are actively working on improving the model right now.

As for the detector, it detects all people in the frame. We pick the one with the highest confidence right now. But you can create a graph that tracks several people. Take a look at how multi-hand tracking is implemented: https://github.com/google/mediapipe/blob/master/mediapipe/graphs/hand_tracking/hand_tracking_desktop_live.pbtxt

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you.

Closing as stale. Please reopen if you'd like to work on this further.

Are you satisfied with the resolution of your issue? Yes No