Problem generating CAMs using the saved model

[Q512147211]

Hi,

I am facing a problem generating CAMs using the saved pre-trained VGG-13 model. The model is loaded correctly and verified using the message "All keys matched successfully." I also printed the input_tensor to verify it was reading the image correctly.

     tensor([[[[ 0.6078,  0.6000,  0.6235,  ...,  0.6314,  0.6157,  0.6314],
      [ 0.6157,  0.6000,  0.5922,  ...,  0.6314,  0.6314,  0.6314],
      [ 0.5216,  0.5137,  0.4824,  ...,  0.6235,  0.6314,  0.6314],
      ...,
      [ 0.3098,  0.2941,  0.2706,  ...,  0.0824,  0.2157,  0.2627],
      [ 0.1294,  0.1608,  0.2471,  ..., -0.1137, -0.0588,  0.1765],
      [ 0.0353, -0.0902, -0.0510,  ...,  0.0902,  0.0431,  0.0667]],

     [[ 0.5137,  0.5137,  0.4902,  ...,  0.5843,  0.5608,  0.5765],
      [ 0.5216,  0.5059,  0.4588,  ...,  0.5843,  0.5765,  0.5765],
      [ 0.4275,  0.4039,  0.3176,  ...,  0.5686,  0.5765,  0.5765],
      ...,
      [-0.4431, -0.4902, -0.5216,  ..., -0.6784, -0.5373, -0.4745],
      [-0.6549, -0.5922, -0.5373,  ..., -0.8902, -0.8275, -0.5843],
      [-0.6941, -0.6863, -0.6941,  ..., -0.7333, -0.7804, -0.7490]],

     [[ 0.6078,  0.5765,  0.6078,  ...,  0.6392,  0.6235,  0.6392],
      [ 0.6157,  0.5765,  0.5922,  ...,  0.6314,  0.6392,  0.6392],
      [ 0.5529,  0.5373,  0.5216,  ...,  0.6157,  0.6392,  0.6392],
      ...,
      [ 0.2392,  0.2000,  0.2000,  ...,  0.0980,  0.2157,  0.2314],
      [ 0.0980,  0.1059,  0.1765,  ..., -0.0588, -0.0431,  0.1529],
      [ 0.0745, -0.0275, -0.0275,  ...,  0.0745,  0.0118,  0.0196]]]])

image_loc = 'test.png'
img = np.array(Image.open(image_loc))
img = cv2.resize(img, (256, 256))
img = np.float32(img) / 255
input_tensor = preprocess_image(img)

targets = None
target_layers = [model.features[-1]]
with GradCAM(model=model, target_layers=target_layers) as cam:
    grayscale_cams = cam(input_tensor=input_tensor, targets=targets)
    cam_image = show_cam_on_image(img, grayscale_cams[0, :], use_rgb=True)
cam = np.uint8(255*grayscale_cams[0, :])
cam = cv2.merge([cam, cam, cam])
images = np.hstack((np.uint8(255*img), cam , cam_image))
Image.fromarray(images)

      An exception occurred in CAM with block: <class 'numpy.exceptions.AxisError'>. Message: axis 2 is out of bounds 
     for array of dimension 0
      ---------------------------------------------------------------------------
      NameError                                 Traceback (most recent call last)
      Cell In[5], line 6
            4     grayscale_cams = cam(input_tensor=input_tensor, targets=targets)
            5     cam_image = show_cam_on_image(img, grayscale_cams[0, :], use_rgb=True)
      ----> 6 cam = np.uint8(255*grayscale_cams[0, :])
            7 cam = cv2.merge([cam, cam, cam])
            8 images = np.hstack((np.uint8(255*img), cam , cam_image))
      
      NameError: name 'grayscale_cams' is not defined

[sivaji123256]

hi @Q512147211 ,Are you able to solve the issue?

[Q512147211]

model = torchvision.models.vgg13(pretrained=False)

for param in model.parameters():
    param.requires_grad = False

model.avgpool = Identity()
model.classifier = nn.Sequential(
    nn.Linear(41472, 100), nn.ReLU(), nn.Linear(100, num_classes)
)
model.to(device)

I was using the pre-trained model as given above and was getting the error due to the

model.avgpool = Identity()

By removing this average pooling layer, I was able to generate CAMs.