Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

inaccurate VQA score #10

Open
Park-ing-lot opened this issue Feb 24, 2023 · 3 comments
Open

inaccurate VQA score #10

Park-ing-lot opened this issue Feb 24, 2023 · 3 comments

Comments

@Park-ing-lot
Copy link

Park-ing-lot commented Feb 24, 2023

When I finetune the model with VQAv2,
It produces a lower score than that proposed in the paper. (43%)

Can you explain where this difference comes from and how to fix it??

@zinengtang
Copy link
Owner

How many epochs did you run. Did you properly initialized the model weights?

@Park-ing-lot
Copy link
Author

10 epochs and I initialized with pre-trained TVLT weights from huggingface named TVLT.ckpt

@irishev
Copy link

irishev commented Mar 5, 2023

Hi! I'm experiencing the same issue and have found no performance difference between randomly initialized models and pre-trained checkpoints.

So, one suspicion is there might be some problems in the initialization. Is there a way to check if the model is properly initialized?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

3 participants