search

simon-ging / coot-videotext

COOT: Cooperative Hierarchical Transformer for Video-Text Representation Learning
Apache License 2.0
288 stars 55 forks source link

Cannot Reproduce Result on paper #49

Closed teang1995 closed 2 years ago

teang1995 commented 2 years ago

Hi, I experimented with your code, but the performance at YouCook2, Yocook2+HowTo100M written in the paper was not reproduced. I experimented with

OS: Ubuntu 20.04 CUDA: 11.2 torch: 1.9.0 + cu111 random seed 1000~1004 config: default

However, as you can see from the picture below, the performance was not reproduced as described in the paper.

image image

Could you tell me more about the settings you used to conduct the experiment? I'd appreciate it if you did that.

Thank you.

teang1995 commented 2 years ago

Oh, I found config for reproduce just now. If performance not reproduced, I will ask you later. Thank you.