guojiajeremy/EDC

EDC

Official PyTorch Implementation of "Encoder-Decoder Contrast for Unsupervised Anomaly Detection in Medical Images".

IEEE Transactions on Medical Imaging 2023. paper

1. Environments

Create a new conda environment and install required packages.

conda create -n my_env python=3.8.12
conda activate my_env
pip install -r requirements.txt

Experiments are conducted on NVIDIA GeForce RTX 3090 (24GB). Same GPU and package version are recommended.

2. Prepare Datasets

Noted that ../ is the upper directory of this folder (EDC). It is where we keep all the datasets by default.

OCT2017

Creat a new directory ../OCT2017. Download ZhangLabData from URL. Unzip the file, and move everything in ZhangLabData/CellData/OCT to ../OCT2017/. The directory should be like:

|-- OCT2017
    |-- test
        |-- CNV
        |-- DME
        |-- DRUSEN
        |-- NORMAL
    |-- train
        |-- CNV
        |-- DME
        |-- DRUSEN
        |-- NORMAL

APTOS

Creat a new directory ../APTOS. Download APTOS 2019 from URL. Unzip the file to ../APTOS/original/. Now, the directory would be like:

|-- APTOS
    |-- original
        |-- test_images
        |-- train_images
        |-- test.csv
        |-- train.csv

Run the following command to preprocess the data to ../APTOS/.

python ./prepare_dataset/prepare_aptos.py --data-folder ../APTOS/original --save-folder ../APTOS

The directory would be like:

|-- APTOS
    |-- test
        |-- NORMAL
        |-- ABNORMAL
    |-- train
        |-- NORMAL
    |-- original

You can delete original if you want.

ISIC2018

Creat a new directory ../ISIC2018. Go to the ISIC 2018 official website. Download "Training Data","Training Ground Truth", "Validation Data", and "Validation Ground Truth" of Task 3. Unzip them to ../ISIC2018/original/. Now, the directory would be like:

|-- ISIC2018
    |-- original
        |-- ISIC2018_Task3_Training_GroundTruth
        |-- ISIC2018_Task3_Training_Input
        |-- ISIC2018_Task3_Validation_GroundTruth
        |-- ISIC2018_Task3_Validation_Input

Run the following command to preprocess the data to ../ISIC2018/.

python ./prepare_dataset/prepare_isic2018.py --data-folder ../ISIC2018/original --save-folder ../ISIC2018

The directory would be like:

|-- ISIC2018
    |-- test
        |-- NORMAL
        |-- ABNORMAL
    |-- train
        |-- NORMAL
    |-- original

You can delete original if you want.

Br35H

Creat a new directory ../Br35H. Go to the kaggle website. Download "yes" and "no". Unzip them to ../Br35H/original/. Now, the directory would be like:

|-- Br35H
    |-- original
        |-- yes
        |-- no

Run the following command to preprocess the data to ../ISIC2018/.

python ./prepare_dataset/prepare_br35h.py --data-folder ../Br35H/original --save-folder ../Br35H

The directory would be like:

|-- Br35H
    |-- test
        |-- NORMAL
        |-- ABNORMAL
    |-- train
        |-- NORMAL
    |-- original

You can delete original if you want.

3. Run Experiments

Run experiments with default arguments.

APTOS

python edc_aptos.py

OCT2017

python edc_oct.py

Br35H

python edc_br35h.py

ISIC2018

python edc_isic.py

Further Improvement

See our new paper "ReContrast: Domain-Specific Anomaly Detection via Contrastive Reconstruction" NeurIPS 2023. It introduces three key elements of contrastive learning into feature reconstruction, i.e., two-view contrastive pair, global similarity, and stop gradient, building a fully 2-D contrastive paradigm. ReContrast also yields SOTA performances on industrial UAD datasets (MVTecAD and VisA).