search

genbing99 / SoftNet-SpotME

Shallow Optical Flow Three-Stream CNN For Macro and Micro-Expression Spotting From Long Videos
45 stars 20 forks source link

Questions about paper #9

Closed xjtupanda closed 1 year ago

xjtupanda commented 1 year ago

Hello! I'm trying to make a picture like Fig.1 in the paper, i.e., teaser.png in SoftNet-SpotME/images. Specifically, how did you make the heat-map/attention-map type pictures in the middle column and on the right? I guess the reddish parts mean higher intensity and the bluish parts indicate weaker. Is it made by placing an intensity mask onto the original picture? I would appreciate it very much if you could share the code. Thanks in advance!

genbing99 commented 1 year ago

Hi, sorry for the late reply. I do not keep the old codes with me, as it was 2 years back.

We use the gradcam to compute the class activation map, then it is placed on the original picture to visualize the activated regions.

xjtupanda commented 1 year ago

Hi, sorry for the late reply. I do not keep the old codes with me, as it was 2 years back.

We use the gradcam to compute the class activation map, then it is placed on the original picture to visualize the activated regions.

Thank you so much! Problem solved.