search

MichiganCOG / A2CL-PT

Adversarial Background-Aware Loss for Weakly-supervised Temporal Activity Localization (ECCV 2020)
MIT License
46 stars 8 forks source link

About hard_index #2

Closed jiujiangluck closed 4 years ago

jiujiangluck commented 4 years ago

I am sorry to intertrupt you. I have a question about hard index. In your paper, the hard index follows: Screenshot from 2020-08-29 16-06-38 But it uses max function in backward in your code. Why?

forward https://github.com/MichiganCOG/A2CL-PT/blob/493fb6e23b60f5a4ba52c596f9c95be86dc33b1a/losses.py#L45-L47

backward https://github.com/MichiganCOG/A2CL-PT/blob/493fb6e23b60f5a4ba52c596f9c95be86dc33b1a/losses.py#L66-L68

kylemin commented 4 years ago

Hi, Thank you for your interest! That is because distmat's are defined differently in the forward and the backward functions. Please refer to lines 39 and 60 (torch.acos and torch.mm are used, respectively). Thank you. Kyle