Jacob Gildenblat

Hi, One way would be to apply image processing on the heatmap: Threshold the heat map, Find it's edge map, Find countours, Take the bounding boxes / center of mass...

If you have a deep learning model, you can add CAM and retrain it, or use grad-cam. Do you have a deep learning model doing your tracking? If so, you...

@MinaRe The images should be in any format that can be read with opencv - i.e .jpg, .png, .pgm etc.

@MinaRe I'm not sure what is the question. One way to load images would to be just use opencv, and load them into numpy arrays. `import cv2` `img = cv2.imread("/home/MinaRe/img.png",...

Thanks for the issue. It seems that .trainable = False freezes the entire model. Also discussed here: https://github.com/fchollet/keras/issues/4510 Can you please elaborate why do you think it should be set...

It can be tuned to your specific problem. As long as you can specify a target cost function, you can use this. I used this for visualizing the PilotNet network...

Notice that the last layer type is target_category_loss, K.sum here is the same as K.max, only a single class is activated.

Flatten-514 won't work as a target layer. It has to be the 2D output of a CNN. So Block8-511 for example or earlier, before the adaptive pooling that turns it...

Hi, 1. They are not class specific any more, because the classifier part is actually discarded when we keep only the embedding part. Instead of being "class" specific, they are...

You are 100% right. I removed the warning in the repository, will update in pypi soon.