Evidence Reconciled Neural Network

Out-of-Distribution detection method.

Data preparation

The ISIC2019 dataset can be download at https://challenge.isic-archive.com/data/#2019

Please change your own path for ISIC2019 dataset in evaluate.py and train.py

p_train_img = '/mnt/mnt_data/ISIC_2019/ISIC_2019_Training_Input'
p_train_label = '/mnt/mnt_data/ISIC_2019/ISIC_2019_Training_GroundTruth.csv'

Code

evaluate.py

To evaluate ID classification and OOD detection:

python evaluate.py

train.py

To train the proposed ERNN on existing dataset

python evaluate.py

dataset.py

Load dataset, split ood categories, split data for five-fold for id categories.

models.py

Implementation for proposed ERNN.

evaluation.py

In line with the work "Out-of-Distribution Detection for Long-tailed and Fine-grained Skin Lesion Images" (https://arxiv.org/abs/2206.15186) in MICCAI 2022.