how to visual the specific channels of feature? about pytorch-grad-cam HOT 10 CLOSED

Hi, I'd like to know this as well! Did you ever figure it out?

from pytorch-grad-cam.

Hi, I'd like to know this as well! Did you ever figure it out?

No, I give up

from pytorch-grad-cam.

@lao-ling-jie can you link source? just trying to better understand what you mean

from pytorch-grad-cam.

@lao-ling-jie can you link source? just trying to better understand what you mean

Thank you very much. For example, the feature map of last resnet block in Resnet50 is 204877 (for input image of 224*224). I want to visualize the first channel of this feature map. How to operate ?

from pytorch-grad-cam.

so by channel are you talking the RGB

from pytorch-grad-cam.

No，it's a specific channel of feature map

from pytorch-grad-cam.

I think I did what you were looking for if I am understanding it correctly - the for loop in line 117 is creating the merged feature map,

therefore the first channel can be accessed here using the following line:

        firstcam = weights[1] * target[1, :, :]

        for i, w in enumerate(weights):
            cam += w * target[i, :, :]

I have zipped my code and you can Ctrl+F firstcam to see what I have done elsewhere in the doc

gradcam_v1.zip

from pytorch-grad-cam.

Thank you so much. I will verify it later.

from pytorch-grad-cam.

@lao-ling-jie verified?

from pytorch-grad-cam.

@lao-ling-jie verified?

I got the corresponding heat map，and think it works.

from pytorch-grad-cam.