How to reproduce your results?

[nasyxx]

Hi GRAM authors,

Thanks for your excellent work.

I'm trying to reproduce your results with your code.

Since I don’t have the VAST27M 150k subset, I tried fine-tuning the MSRVTT dataset using the three checkpoints based on VAST model, 4modal, and 5modal to reproduce your results. However, regardless of which dataset or which base checkpoint I use, I cannot reach your reported 64% R@1.

Could you let me know on which base model you fine-tuned the MSRVTT dataset?

Here are my reproduced results:

base model	T2D r1	D2T r1
VAST model	59	59
4modal	60.5	61
5modal	60.6	60.9