Skip to content

Latest commit

 

History

History
49 lines (36 loc) · 717 Bytes

Readme.md

File metadata and controls

49 lines (36 loc) · 717 Bytes
Raw

Interpretability Metrics

Setup

  1. Put ImageNet validation dataset under "data/imagenet/val" folder
  2. Install libraries
# For PIP
pip install -r requirements.txt

# For Conda
conda install --file requirements.txt -y

Run

Degradation

# Evaluation Example
python3 scripts/degradation.py Saliency GradCam -m resnet50 -s results/ -d cuda:1

# Draw Graph Example
python3 scripts/graph.py -s results -o ./result.png

Supported Methods

  1. Captum
  • Saliency
  • GuidedBackprop
  • Deconvolution
  • Occlusion
  • LayerGradCam
  • GuidedGradCam
  • IntegratedGradients
  1. Custom Captum
  • OnOffCam
  • GuidedOnOffCam

Supported Models

  • AlexNet
  • VGG16
  • ResNet18
  • ResNet50
  • GoogLeNet