Confused about the CAM method?

[jokingww]

Nice Work. In the code you get the cam_map using only one Conv2d layer. According to my intuition, here should not use a method like GradCAM to get CAM. And I also see this process in other place. Can you explain why this this works？ Thank you.

[rajeevyasarla]

in this code I am computing CAM using https://github.com/zhoubolei/CAM

for high quality CAM map you can use GradCAM method, https://github.com/jacobgil/pytorch-grad-cam

the following code will produce CAM ma using GradCAM method

from pytorch_grad_cam import GradCAM, HiResCAM, ScoreCAM, GradCAMPlusPlus, AblationCAM, XGradCAM, EigenCAM, FullGrad
from pytorch_grad_cam.utils.model_targets import ClassifierOutputTarget
from pytorch_grad_cam.utils.image import show_cam_on_image
from torchvision.models import resnet50

d0, d1, d2, d3, d4, d5, d6,  pred_edges, cam_map, bag_map, pred_class = dino.student_encoder(inputs_v)
target_layers = [ dino.student_encoder.class_input]
input_tensor = inputs_v
targets = [ pred_class]

with GradCAM(model= dino.student_encoder, target_layers=target_layers) as cam:
  grayscale_cam = cam(input_tensor=input_tensor, targets=targets)
  grayscale_cam = grayscale_cam[0, :]