search

Finspire13 / CMCS-Temporal-Action-Localization

Completeness Modeling and Context Separation for Weakly Supervised Temporal Action Localization (CVPR2019)
MIT License
150 stars 17 forks source link

late-fusion #12

Closed dsj1840923452 closed 5 years ago

dsj1840923452 commented 5 years ago

Hi, excuse me, I recently ran through the code you provided. According to the I3D features in the paper, you finally adopted late-fusion, but I detected that the flow alone is even higher than that of late-fusion. What's the reason?thank you

Finspire13 commented 5 years ago

Yes, we also found that results of only flow and late-fusion are similar for I3D (see supplementary). Maybe you can try smaller fusion weight for RGB.

dsj1840923452 commented 5 years ago

Yes, we also found that results of only flow and late-fusion are similar for I3D (see supplementary). Maybe you can try smaller fusion weight for RGB.

Ok, thanks for your reply

dsj1840923452 commented 5 years ago

Yes, we also found that results of only flow and late-fusion are similar for I3D (see supplementary). Maybe you can try smaller fusion weight for RGB.

Ok, thanks for your reply

Hello, another question, is the mAp of the training set and the result of the test set similar, is that right?thank you