Can you produce a dynamic auto-cropped video?

Hi

I was wondering if it is/would be possible to produce a non labeled video from the dynamic autocropping frames. The reason is I have been trying to use this:

to automatically recognize behavior in fish, but it would be desirable to limit video around the animal, as the algorithm is sensitive to any movement.

thanks!

You can, just save the bounding box coordinates (https://github.com/AlexEMG/DeepLabCut/blob/master/deeplabcut/pose_estimation_tensorflow/predict_videos.py#L434) or the frame wise data and then make a video.