This repository contains the software developed at FlySleepLab (Liu Lab) @VIB-KU Leuven CBD to power behavioral tracking in Drosophila melanogaster. It consists of a collection of open source software either published by other groups (like DeepLabCut or by ourselves.
- Try training with less labeled points i.e. one point or feature to be tracked.
Check https://github.com/VIBFlySleepLab/FlyRecognition/blob/master/DLC/examples/Fly-Sayed-2019-02-20/videos/outpy05_reshapeDeepCut_resnet50_FlyFeb20shuffle1_200_labeled.mp4 to see how DLC does tracking the separator of the arena. It's a trivial task, but it works.
-
Label easier objects i.e. the corners of the arena, which don't change
-
Crop videos to focus just on the relevant ROI
Check https://github.com/VIBFlySleepLab/FlyRecognition/blob/master/DLC/examples/Track_flies-Sayed-2019-02-20/videos/outpy05_reshaped_topDeepCut_resnet50_Track_fliesFeb20shuffle1_200_labeled.mp4 and https://github.com/VIBFlySleepLab/FlyRecognition/blob/master/DLC/examples/Track_flies-Sayed-2019-02-20/videos/outpy05_reshaped_bottomDeepCut_resnet50_Track_fliesFeb20shuffle1_200_labeled.mp4 to check how well DLC does after 200 iterations looking at each area separately. Mind that DLC was trained on a different video (outpy04.avi) i.e. it's predicting based on the frames of another video (it's however, very similar).
-
Add more videos
-
Start something out of DLC (by ourselves)