https://github.com/yuneg11/interpretability-metrics

Interpretability Metrics
https://github.com/yuneg11/interpretability-metrics

captum degradation interpretability metrics pytorch

Last synced: about 2 months ago
JSON representation

Interpretability Metrics

• Host: GitHub
• URL: https://github.com/yuneg11/interpretability-metrics
• Owner: yuneg11
• License: mit
• Created: 2020-08-03T07:15:18.000Z (almost 5 years ago)
• Default Branch: master
• Last Pushed: 2022-01-14T19:49:41.000Z (over 3 years ago)
• Last Synced: 2025-02-15T02:20:09.542Z (4 months ago)
• Topics: captum, degradation, interpretability, metrics, pytorch
• Language: Python
• Homepage:
• Size: 25.4 KB
• Stars: 0
• Watchers: 2
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: Readme.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# Interpretability Metrics

## Setup

1. Put ImageNet validation dataset under "data/imagenet/val" folder

2. Install libraries

```bash

# For PIP

pip install -r requirements.txt

# For Conda

conda install --file requirements.txt -y

```

## Run

### Degradation

```bash

# 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

2. Custom Captum

- OnOffCam

- GuidedOnOffCam

#### Supported Models

- AlexNet

- VGG16

- ResNet18

- ResNet50

- GoogLeNet