How to combine the score of CoViAR with the score of TSN?

[ZealACMer]

After obtained the score of CoViAR, how to combine the score with TSN by late fusion? Is there any file I can use to complete this process, thank you very much.

[chaoyuaw]

Hi, I'm not sure if I understand your question correctly. After training a model, we run test.py and test with 25 segments (https://github.com/chaoyuaw/pytorch-coviar/blob/master/test.py#L28). In a sense this is like a TSN-style testing. Is this the "TSN" you're referring to? Or do you want to combine the score of CoViAR model with a TSN model you obtained elsewhere?

[ZealACMer]

I am sorry for not making my question clear enough. What I mean is as shown in the pic below: How did you combine the temporal-stream network using 7 segments with BN-inception[14] with your model by late fusion? Is it a score fusion? Is there any file I can use to do such score fusion? Thank you so much for your attention. And looking forward to your reply.

[chaoyuaw]

Thanks for clarifying! The part of the code isn't included in this repo, but it's doing the same thing as combine.py Namely, it's score fusion.

[ZealACMer]

Thank you very much for your prompt reply, so can I use combine.py in this repo to do the score fusion with respect to CoViAR + flow, or I need to write the file by myself? Can you add the relevant file in this repo please?

[chaoyuaw]

Hi sorry for the late reply. This repository is a reimplementation, and the original code I used to combine CoViAR and Flow isn't compatible with this repo. So unfortunately I don't have any code at hand that I can share. I think you'll have to modify combine.py yourself for this.