dbstjswo505 / squidnet Goto Github PK

Are you planning to share the pretrained model weights?
Thank you.

Hello，as shown in the code，the model directly reads BMR results of TVR dataset and then do the training.
If I want to do it in ActivityNet ,can you release the ActivityNet BMR result or just the BMR model trainging code? So that I can reproduce for other dataset.
Thanks a lot!

Hello! Thank you for your great work and sharing. I love your insightful idea.

I read the code, and I have a question about BMR.

It seems that training & inference phase utilize pre-extracted BMR prediction results,
but I couldn't find the code for training the BMR module & extract the result.

Could you provide the code or a checkpoint? or could you explain how did the BMR module was trained & how did the model architecture look & which loss was used? (as training BMR only may be hard to use moment de-biasing loss)

Thank you for your consideration :)

Hello! Can you publish the code for the ActivityNet dataset？

could you release the performance of TVR (val) not test-public , because The website for obtaining the performance of the TVR dataset has been closed

hi, can you upload or send me the key part to reproduce the results?

Hi, thanks for your excellent work!

When reading the code, I found it is not consistent to your model in the paper,
I'm wondering if there any detailed descriptions or interpretations for such differences, or if I misunderstood the code, could you please update mode interpretations so that I can learn from this as a beginner.
Thank you very much if you can reply!

Does this framework support multiple GPUs training?