What I understand is that the input image size is 224x224, with a patch size of 8, so the final visualization would be 28x28. However, it is clear that the weight score map here is not 28x28 in size. I would like to know how to obtain a visualization result as shown in the image, and how the value of each pixel is calculated.
What I understand is that the input image size is 224x224, with a patch size of 8, so the final visualization would be 28x28. However, it is clear that the weight score map here is not 28x28 in size. I would like to know how to obtain a visualization result as shown in the image, and how the value of each pixel is calculated.