https://github.com/jusiro/constrained_anomaly_segmentation

[MedIA 2022] Constrained unsupervised anomaly segmentation.
https://github.com/jusiro/constrained_anomaly_segmentation

Last synced: 14 days ago
JSON representation

[MedIA 2022] Constrained unsupervised anomaly segmentation.

• Host: GitHub
• URL: https://github.com/jusiro/constrained_anomaly_segmentation
• Owner: jusiro
• License: mit
• Created: 2022-05-25T13:20:42.000Z (over 2 years ago)
• Default Branch: main
• Last Pushed: 2023-03-28T00:27:35.000Z (over 1 year ago)
• Last Synced: 2024-05-20T03:18:51.395Z (6 months ago)
• Language: Python
• Homepage:
• Size: 815 KB
• Stars: 17
• Watchers: 1
• Forks: 1
• Open Issues: 1
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

• Unsupervised-Anomaly-Detection-in-Brain-MRI - [Code

README

        
# Constrained Unsupervised Anomaly Segmentation of Brain Lesions

This repository contains code for unsupervised anomaly segmentation in brain lesions. Specifically, the implemented methods aim to constrain the optimization process to force a VAE to homogenize the activations produced in normal samples.

  

If you find these methods useful for your research, please consider citing:

**J. Silva-Rodríguez, V. Naranjo and J. Dolz, "Looking at the whole picture: constrained unsupervised anomaly segmentation", in British Machine Vision Conference (BMVC), 2021.** [(paper)](https://www.bmvc2021-virtualconference.com/assets/papers/1011.pdf)[(conference)](https://www.bmvc2021-virtualconference.com/conference/papers/paper_1011.html)

**J. Silva-Rodríguez, V. Naranjo and J. Dolz, "Constrained unsupervised anomaly segmentation",  Medical Image Analysis, vol. 80, p. 102526, 2022.** [(paper)](https://www.sciencedirect.com/science/article/pii/S1361841522001736)

## GRADCAMCons: looking at the whole picture via size constraints

```

python main.py --dir_out ../data/results/gradCAMCons/ --method gradCAMCons --learning_rate 0.00001 --wkl 1 --wae 1000 --t 10

```

## AMCons: entropy maximization on activation maps

```

python main.py --dir_out ../data/results/AMCon/ --method camCons --learning_rate 0.0001 --wkl 10 --wH 0.1

```

## Visualizations







## Contact

For further questions or details, please directly reach out Julio Silva-Rodríguez ([email protected])