search

showlab / EgoVLP

[NeurIPS2022] Egocentric Video-Language Pretraining
https://arxiv.org/pdf/2206.01670.pdf
220 stars 19 forks source link

Questions about preprocess for downstream tasks #18

Closed dreamerlin closed 1 year ago

dreamerlin commented 1 year ago

Thanks for your great work.

I would like to use your repo (https://github.com/EGO4D/episodic-memory) for downstream tasks. So, I want to ask some questions about the pre-process for these tasks. Should I also resize and chunk the clips (provided by the official Ego4d) same as the EgoVLP?

QinghongLin commented 1 year ago

Hi! If you want to use the checkpoint to extract features for NLQ, yes, please try to chunk and resize them as 256 (short size). btw, you can directly download the nlq & mq features have provided in the repo.

dreamerlin commented 1 year ago

Seemingly no need to chunk