search

linjieli222 / HERO

Research code for EMNLP 2020 paper "HERO: Hierarchical Encoder for Video+Language Omni-representation Pre-training"
https://arxiv.org/abs/2005.00200
MIT License
230 stars 34 forks source link

TVC cannot reproduce results #28

Closed Liu0329 closed 3 years ago

Liu0329 commented 3 years ago

My cider result is around 35, far from 49.98 in the paper. Anyone has reproduced the results ?

linjieli222 commented 3 years ago

@Liu0329

We have run several experiments to reproduce the results after the release. Here is the log of a reproduced experiment:

https://convaisharables.blob.core.windows.net/hero/finetune/tvc_log.txt

On val split, we were able to get similar performance as what we reported in the paper. Please let me know how I can help you with this.

Liu0329 commented 3 years ago

That looks good. I use larger batch -size and 4 gpus. I will try your config again.