collate_fn for dataloader and extract vision features

Segment long videos and multithreading in EnTubeDataset

Analyze running time of process_images() and process_video_frames()

Remove HF home

inspect VideoReader time

Debugging learn process_images

Restructure repo and refine annotation

Fix pickle file name

Add memory snapshot

Debug EnTubeDataset

I think batching data loader is more general: give batch_size=1 in case limited mem

Back to multithreading for individual sample

Debug dataloader

debug process_images

debug process_images

debug process_images

break

Debug process_images

Debug process_images

Fix impor

Rename dataset.py to entube_dataset.py

Update code

remove self arg

add import json in preprocessor.py

Add config_file arg

Add import torch in preprocessor.py

Add preprocessing module to pre-extract vision features