MST-MIXER/custom_datasets/README.md
2024-07-08 11:41:28 +02:00

1.2 KiB

  1. Download the raw Charades train/val data
  2. Download the raw Charades test data
  3. Install SAM
  4. Segment the frames
    python segement.py --sam_ckpt path_to_sam_ckpt --avsd_root path_to_charades_trval_frames --crop_root path_to_save_the_trval_crops  --mode segment --start start_idx --end end_idx
    python segement.py --sam_ckpt path_to_sam_ckpt --avsd_root path_to_charades_test_frames --crop_root path_to_save_the_test_crops  --mode segment --start start_idx --end end_id
    
  5. Embed the crops
    python segement.py --sam_ckpt path_to_sam_ckpt --crop_root path_to_save_the_trval_crops  --mode emebed --embed_root ../features/sam  --start start_idx --end end_idx
    python segement.py --sam_ckpt path_to_sam_ckpt --crop_root path_to_save_the_test_crops  --mode emebed --embed_root ../features/sam_testset  --start start_idx --end end_idx
    
    
  6. Preprocess and log the data
    python dataset.py --split train
    python dataset.py --split val