fmu2 / snag_release

Official Implementation of SnAG (CVPR 2024)
21 stars 2 forks source link

Question about replication. #4

Open ZijiaLewisLu opened 4 weeks ago

ZijiaLewisLu commented 4 weeks ago

Hi,

Thanks for your interesting work and releasing the code.

I am trying to replicate the model on Ego4D-NLQ. However, the performance of R5 is lower than your released weights and scores in paper, while R1 is higher. Results are shown below. Can you let me know how to replicate the model weight your released (e.g., what is your seed, batch size and number of gpu)?

image

Thanks! Zijia

fmu2 commented 3 weeks ago

Thanks for your interest. The full training config can be found in the released checkpoint folder. Please make sure to train with the provided annotation file.