Results 117 comments of Eyal Gruss

and the thing is that normalizing the gradients will always highlight something. but for finding negative evidence the magnitude is important

also: grads = grads / (np.max(grads) + K.epsilon()) in grad cam

@keisen so in the paper: guided grad cam = visualize_saliency(backprop_modifier='guided') * visualize_cam() this is different from the "guided grad cam" implied in keras-vis: visualize_cam(backprop_modifier='guided') you can see an example here:...

@thibtld thanks. i can confirm gradient_override_map now works for me with tf 1.15

@gaikwadrahul8 thanks for your kind followup. i can confirm that the issue was solved when going from firefox 107.0.1 (broken) to 108.0 (fixed; dec 2022) i tested with both tfjs...

fix here: https://github.com/eyaler/impersonator and colab: https://colab.research.google.com/github/eyaler/avatars4all/blob/master/ganozli.ipynb

notebook is currently broken due to changes of ubuntu/python versions and its compatibility with the project's cuda and pytorch versions. it getting hard to keep up with google constantly changing...

thanks! for now i am just doing something like: ``` sa = divsufsort(np.array([ord(c) for c in text])) trans = ''.join(text[sa[i] - 1] for i in range(len(sa))) index = list(sa).index(0) ```...

original pix2pix implementation does not have bn on last convolution