- Skeleton video: deeplabcut → .h5 → script → .mp4, in
/train_videofolder. - In
/train_datafolder, CSV files, three cols: start, end, duration. Hand labeled grooming moments. - Initialize/config model (critical, many magic numbers, all
about tensorflow & keras), then train model, save to a
.h5 - Output prediction CSV
- (optional) Our decision model: when to mark 1 or 0
- plot based on decision output or prediction CSV