Jacob Gildenblat

This is really great and will bring CAM in object detection to a large audience. Do you think there is a way of also getting gradients there for GradCAM (and...

Hi Jeffery, Can you please explain what this PR does. The code already supports batches.

Hi, Sorry for the very late reply. And thanks for this PR! going to merge this ! Was curious if you have examples of test time augmentations that worked in...

Hi, thanks a lot, this is impressive. I'm looking into this, might take me about a week. There seem to be some minor bugs: --roimode 2 returns me an error,...

The code in the repo saves the entire model with pickling, instead of the state dict, which is actually a bad practice. A better way would be to save only...

Still relevant? Try printing x.size() in forward. It should be [batch_size, 3, height, width]

The gradient is the gradient of the output with respect to each one of the activation outputs. Therefore the gradient shape is the same as the activation outputs shape.

Hi, Can you please try playing with the parameters: -- lr 0.0001 (larger than default) or -- lr 0.000001 (smaller than default) and also trying running for more iterations: --num_batches...

Also realized now you were running on the lincoln image after x2 resizing. The LR lincoln image I produced the result with is this one: http://www.wisdom.weizmann.ac.il/~vision/zssr/real/img_002_SRF_2_LR.png I'd suggest running on...

Hi, I'm interested in this too! I'm trying running an extremely light weight model (on CPU), the forward pass takes 10ms. I actually got to the same bottleneck code. So...