CAMEL for Retina OCT Image Classification and Segmentation [WACV 2025]

This is the official implementation of "CAMEL: Confidence-Aware Multi-task Ensemble Learning with Spatial Information for Retina OCT Image Classification and Segmentation" accepted in IEEE/CVF Winter Conference on Applications of Computer Vision (WACV 2025)

News

🔥October 2024, Paper accepted at WACV 2025 🎉.

Requirements

• Python 3.10
• torch 2.1.2
• TensorFlow 2.9.1
• segmentation-models 1.0.1

Data Preparation

For training and testing the models, you can use the public dataset OCT5K

The dataset covers semantic segmentation and object detection tasks.

• "Images": Original OCT Images
• "Masks": pixel-wise annotations with three manual gradings for 1,672 images and 2,924 masks with single automatic grading
• "Detection": CSV files for object detection labels

Training

To train the network, you can run the following command:

python3 train.py --batch_size 4 --aug 5 --model resnet101 -img_size 320 -erm_weight 0.2 --ece_weight 0.01

Testing on Dataset

After training, you can use test.ipynb for performance evaluation and visual inference. Open the notebook and follow the instructions to evaluate the trained model on your dataset.

Code Description

• train.py: Code for training CAMEL
• test.ipynb: Code for testing CAMEL

utils

• loss.py: Code for loss functions
• utils.py: util functions, including labeling, preprocessing codes

preprocessing

• augmentation.py: Code for image augmentation
• image_to_mask.py: Applies our new OCT image preprocessing method, converting processed annotation images into the .npy format.

Citation

@inproceedings{
jung2025camel,
title={CAMEL: Confidence-Aware Multi-task Ensemble Learning with Spatial
Information for Retina OCT Image Classification and Segmentation},
author={Juho Jung, Migyeong Yang, Hyunseon Won, Jiwon Kim, Jeongmo Han, Joonseo Hwang, Daniel Duck-Jin Hwang, and Jinyoung Han},
booktitle={},
year={2025},
url={}
}