I’m currently trying to use DLC2.2 to track the movements of 4 mice in a frame, where 2 of them are in one cage, and the other two in another. I’ve also shaved the fur off of the back of one mouse from each cage, to make it easier to distinguish between the animals. I’ve currently trained it with 240 frames from different videos at a frame size of 480x360. The problem is that after I evaluate the network, it seems like DLC has double labelled some of the mice so it looks like there are more than 4 “body” and more than 4 “head” markers in some of the frames. How many more frames should I add to train the network?
Hey @UTstudent! Is that double-labeling occurring right after network evaluation (i.e., verified with
create_video_with_all_detections) or after the conversion to tracklets (visible in the GUI)? Have you run the cross-validation?
It happened after I did the network evaluation and ran the cross-validation.