yulongwang12 / visual-attribution

Pytorch Implementation of recent visual attribution methods for model interpretability

Pytorch Visual Attribution Methods

A collection of visual attribution methods for model interpretability

Including:

• Vanilla Gradient Saliency
• Grad X Input
• Integrated Gradient
• SmoothGrad
• Deconv
• Guided Backpropagation
• Excitation Backpropagation, Contrastive Excitation Backpropagation
• GradCAM
• PatternNet, PatternLRP
• Real Time Saliency
• Occlusion
• Feedback
• DeepLIFT
• Meaningful Perturbation

Setup

Prerequisities

• Linux
• NVIDIA GPU + CUDA (Current only support running on GPU)
• Python 3.x
• PyTorch version == 0.2.0 (Sorry I haven't tested on newer versions)
• torchvision, skimage, matplotlib

Getting Started

• Clone this repo:

git clone [email protected]:yulongwang12/visual-attribution.git
cd visual-attribution

• Download pretrained weights

cd weights
bash ./download_patterns.sh  # for using PatternNet, PatternLRP
bash ./download_realtime_saliency.sh # for using Real Time Saliency

Note: I convert caffe bvlc_googlenet pretrained models in pytorch format (see googlenet.py and weights/googlenet.pth).

Visual Saliency Comparison

see notebook saliency_comparison.ipynb. If everything works, you will get the above image.

Weakly Supervised Object Localization

TBD

Citation

If you use our codebase or models in your research, please cite this project.

@misc{visualattr2018,
  author =       {Yulong Wang},
  title =        {Pytorch-Visual-Attribution},
  howpublished = {\url{https://github.com/yulongwang12/visual-attribution}},
  year =         {2018}
}

visual-attribution's Issues

Code for computing patterns?

Hi,

thanks for sharing this code!
Could you add some code for computing the patterns for PatternNet etc.? That would be very helpful.

Thanks,
Robin

Pattern weights

Hi,

the download URL seems to have gone.
Is there a copy somewhere else?
Thank you!

Best regards

Thomas

Any interest for supporting LayerCAM

Hi, @yulongwang12 ,
Our paper "LayerCAM: Exploring Hierarchical Class Activation Maps for Localization" is accepted by TIP recently, which can visualize the class activation maps from any CNN layer of an off-the-shelf network. Could you add our method to your popular repository for more people to try this method? Our method is a simple modification of Grad-CAM. It should easy to implement. Here is the paper and code. Hope for your reply.

false

The link to the weight is out of date

How to apply this to any classifier?

Hi Thank you for this contribution.

I am wondering if you have example code for how to apply this to classifiers that are not resnet-50, inception-v3, or vgg16? I am hoping to get excitation backprop figures for a 3 layer cnn and a 3 layer fully connected network.

Thanks for the great work!

Add License to Repo