search

ivanorsolic / oricodes

A website documenting my self-driving RC car project/Master's thesis and my ML/DL cookbooks. Built with Hugo, Grav and the Learn theme. Contains video demos, code, schematics, 3D models and Baby Yoda pics drawn by Van Gogh and Picasso.
https://ori.codes
10 stars 0 forks source link

Visualization: Saliency Maps :: Ori Codes #5

Open utterances-bot opened 3 years ago

utterances-bot commented 3 years ago

Visualization: Saliency Maps :: Ori Codes

Ori's self-driving RC car

https://ori.codes/artificial-intelligence/visualization/saliency-maps/

gcordova19 commented 3 years ago

Hi,

I run your code but I didn't know why you use
upscaled_activation = np.ones((53, 73))

Checking your inspiration github i realize
upscaled_activation = np.ones((3, 6))

So I have problems averaged_activation = np.mean(activations[layer], axis=3).squeeze(axis=0) * upscaled_activation In my case np.mean(activations[layer], axis=3).squeeze(axis=0) is (1, 18) if I multiply by upscaled_activation in any case work by dimentions.

I would like to know why to use that dimension in upscaled_activation