Question about the Figure 5 (CA visualization) in paper

myungsub / CAIN

Source code for AAAI 2020 paper "Channel Attention Is All You Need for Video Frame Interpolation"

MIT License

320 stars 43 forks source link

Question about the Figure 5 (CA visualization) in paper #13

Open Lecanyu opened 3 years ago

Lecanyu commented 3 years ago

Hi, I'm interested about the Figure 5, Visualization of internal feature maps with their channel attentions, in your paper.

Could you please tell me more details about how you calculate the activation maps for channels? Or pubilish the relevant code about the activation map calculation.

Thanks

issakh commented 2 years ago

Hi, did you manage to figure this out?

fiftywu commented 2 years ago

I also want to know it. wu hu ~

myungsub commented 2 years ago

Sorry for the late reply. From what I remember, I simply grabbed the channel with the highest channel attention score for each Residual Group and displayed it as a heatmap (don't exactly remember the color coding of the heatmap, though).

For instance, for ResGroup1, there are 12 ResBlocks and 128 channels, so 128*12 attention scores in total. I just took the one with the maximum attention score. Same for the other Groups.