search

facebookresearch / LaViLa

Code release for "Learning Video Representations from Large Language Models"
MIT License
491 stars 46 forks source link

The pre-training weights of Dual-Encoder Baseline (with TSF-B/L) #20

Closed daiguangzhao closed 1 year ago

daiguangzhao commented 1 year ago

Thanks so much to the authors for their contributions to the egocentric community! I wanted to use your excellent code as my codebase in my work, but after I removed the pre-training weights (clip_openai_timesformer_base.narrator_rephraser.ep_0005.md5sum_d73a9c.pth), the performance of epic100 became very bad. Therefore, would you mind providing the pre-training weights for Dual-Encoder Baseline (with TSF-B/L)? Thanks a lot!

zhaoyue-zephyrus commented 1 year ago

Hi @daiguangzhao ,

Please take a look at MODEL_ZOO.md#narrator. The second column from the right should hopefully work.

Best, Yue

daiguangzhao commented 1 year ago

thanks for your reply! Thanks again for your excellent contribution to the ego community.