Skip to content

Latest commit

 

History

History
executable file
·
22 lines (13 loc) · 645 Bytes

README.md

File metadata and controls

executable file
·
22 lines (13 loc) · 645 Bytes

MEME - MoE for Multimodal Egocentric Videos

experimenting with egocentric

Preparation

Dataset

Download the dataset from the [official webpage][ego4d_page] and place them in data/ folder. Specify the paths in config.yaml

Run the preprocessing script using:

python utils/get_clip_features.py -c config.yaml

This creates JSON files in data/dataset/nlq_official_v1 that can be used for training and evaluating the VSLNet baseline model.

Video features

Download the official video features released from [official webpage][ego4d_page] and place them in data/features/nlq_official_v1/video_features/ folder.