search

utkuozbulak / pytorch-cnn-visualizations

Pytorch implementation of convolutional neural network visualization techniques
MIT License
7.81k stars 1.49k forks source link

Support for Score-CAM #76

Closed haofanwang closed 4 years ago

haofanwang commented 4 years ago

Any interest in implementing Score-CAM which is a gradient-free extension of CAM. Here is the reference code and paper. It should easy to implement with some modifications on Grad-CAM code.

utkuozbulak commented 4 years ago

Hey, it is already implemented in the linked repository, why would it need another implementation?

haofanwang commented 4 years ago

Hi, @utkuozbulak. The original code may not be easy to understand after encapsulation. Your project has implemented many common methods in the same style, others may be able to better compare the differences between this method and different CAM. I'm willing to implement and PR if it is ok for you.

utkuozbulak commented 4 years ago

Hey, I did not realize you were the author, hah. Sure thing, send a PR! I hope putting your approach here increases its reach as well as its adoption rate!

Can you please format it to be a single python file, you can also use the functions in misc_functions.py if they are any useful. Also, please mention that you are the author in the file. For example, see: https://github.com/utkuozbulak/pytorch-cnn-visualizations/blob/master/src/generate_regularized_class_specific_samples.py