Gradcam for models that return a single value

[wyli]

Discussed in #5528

^{Originally posted by rajanish4 November 15, 2022}
Hi there,

I am currently running Grad cam class for my binary classification model that returns a single value. I believe, the grad cam compute function requires multiple class labels for it to work.
So, i would like to request a modification of this that can be used for such binary classification tasks.

Thanks

(the class score function and the subclasses are not compatible with such input models

MONAI/monai/visualize/class_activation_maps.py

Lines 125 to 126 in 5e6f105

	def class_score(self, logits, class_idx):
	return logits[:, class_idx].squeeze()

)

[thatgeeman]

Hello 👋 I'd like a go at it if no one else is working on this 🤗

[wyli]

sure, please go ahead, let me know if you need any help

[thatgeeman]

Finally got some time to work on this. Not sure if I've mistaken the intent here, but I'm not able to reproduce the issue. For a binary classifier, I'm setting the out_channels=1 below and everything seems to work.

from monai.networks.nets import DenseNet121
from monai.visualize import GradCAM 
model_2d = DenseNet121(spatial_dims=2, in_channels=1, out_channels=1)
cam = GradCAM(nn_module=model_2d, target_layers="class_layers.relu")
result = cam(x=torch.rand((1, 1, 64, 64)))

[wyli]

thanks for looking into this @thatgeeman, indeed it's already supported. I'm closing this and please help update one of the test cases for the single channel output if you are interested...

MONAI/tests/test_vis_gradcam.py

Line 103 in 7614aca

def test_shape(self, cam_class, input_data, expected_shape):