I am running another video and in the label frames toolbox, and I have noticed that deeplabcut has only extracted 19 frames. We went back to the video and we were able to single out 47 frames total that the rabbit is in. I looked at the video on YouTube and that example only has 19 frames also. Does deeplabcut only extract 19 frames regardless? If not, how can I get closer to the 47 frames that we found in our video? We need to analyze all 47 frames because even though joint angles are changing slowly, all the frames are needed to determine how the joint angles are changing over time. I’m not sure if the software deletes redundant frames, because I do see at the img.png part at the top of the label frames toolbox is skipping numbers. Any help is greatly appreciated!
Hi @jpierce, the default is to extract 20 frames; this can be seen in your project config.yaml file at
numframes2pick. Why would you want to extract all frames from your video though? You should only extract a relatively small number of representative ones, and let the model learn from those such that predictions can then be made over the entire video. You write that you’re running another video: does that mean that a model was already trained on another one? If so, have you tried to analyze the new one directly and see whether it generalizes well?
Sorry for the late response! I was just asking for clarification for my PI. And I just meant that by running another video, I was doing another rabbit video after the first one that I did. Thanks for your help!