search

zhipeng-wei / TT

13 stars 2 forks source link

About the Figure2(c) in the paper #2

Open lwmming opened 2 years ago

lwmming commented 2 years ago

Hi, Zhipeng. I have an issue about the Figure2(c) in the original paper. How to perform Grad-CAM for each frame in a video? For the video models, such as NL-101, NL-50,TPN-101,TPN-50, the temporal dimension will decrease (32->16->8->4...) as the network forward. Thank you for your reply.

zhipeng-wei commented 2 years ago

Hi, Iwmmming. It also directly utilizes torch.nn.functional.interpolate to match features with the temporal dimension.

lwmming commented 2 years ago

Ok, I know, thank you~