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