Analyze videos on cropped region, then create tracklets, etc. of full video

using dlc 2.2b8 (CPU) on an iMac running Catalina

Everything’s working great, but when I run deeplabcut.analyze_videos (on a different machine with GPU) with cropping=True, the points are always offset when I make tracklets or run create_video_with_all_detections. I understand this isn’t really “cropping”, just analyzing a subset of points in the full video.

I tried actually cropping before running analyze_videos, but the video quality is significantly degraded. Being able to analyze just a region of the full video frame is a huge timesaver for me.

It looks like the cropping parameters are stored in the “_meta.pickle” file. Has there been any thought to using the metadata file to re-align the detected points to the original video?

Thanks,
Dan