https://github.com/ErikEnglesson/GJS

The official code for the NeurIPS 2021 paper Generalized Jensen-Shannon Divergence Loss for Learning with Noisy Labels (https://arxiv.org/abs/2105.04522)
https://github.com/ErikEnglesson/GJS

Last synced: 7 days ago
JSON representation

The official code for the NeurIPS 2021 paper Generalized Jensen-Shannon Divergence Loss for Learning with Noisy Labels (https://arxiv.org/abs/2105.04522)

• Host: GitHub
• URL: https://github.com/ErikEnglesson/GJS
• Owner: ErikEnglesson
• Created: 2021-10-26T07:10:36.000Z (about 3 years ago)
• Default Branch: main
• Last Pushed: 2021-12-29T18:29:46.000Z (almost 3 years ago)
• Last Synced: 2024-08-02T15:31:01.384Z (3 months ago)
• Language: Python
• Homepage:
• Size: 47.9 KB
• Stars: 21
• Watchers: 1
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# Generalized Jensen-Shannon Divergence Loss for Learning with Noisy Labels

The official code for the NeurIPS 2021 paper [Generalized Jensen-Shannon Divergence Loss for Learning with Noisy Labels](https://arxiv.org/abs/2105.04522) 

## Environment Setup

Create conda environment, activate environment, and install additional pip packages 

```bash

conda env create -f gjs_env.yml -n gjs

conda activate gjs

python -m pip install -r requirements.txt

```

## Running Experiments

Please check scripts/ folder for yaml files corresponding to different experiments.

For example, to run JS on 40% symmetric noise on the full CIFAR-10 training set, run the following

```bash

python train.py -c scripts/C10/sym/js-40.yaml \

                --data_dir /path/to/dataset/

```

or GJS on 20% asymmetric noise on CIFAR-100

```bash

python train.py -c scripts/C100/asym/gjs-20.yaml \

                --data_dir /path/to/dataset/

```

or GJS on WebVision

```bash

python train.py -c scripts/WebVision/gjs.yaml \

                --data_dir /path/to/dataset/

```