ICCV_2023_CVAMD [Order-ViT]

Order-ViT: Order Learning Vision Transformer for Cancer Classification in Pathology Images

Description

The overall model architecture is as follows. Categorical classification and sequential relationship classification problems are performed simultaneously.

Datasets

All the models in this project were evaluated on the following datasets:

• Colon_KBSMC (Colon TMA from Kangbuk Samsung Hospital)
• Colon_KBSMC (Colon WSI from Kangbuk Samsung Hospital)
• Gastric_KBSMC (Gastric from Kangbuk Samsung Hospital)

How to run

Install dependencies

# clone project
git clone https://github.com/Leejucheon96/ICCV_2023_CVAMD-Order-ViT-.git


# [OPTIONAL] create conda environment
conda env create -f order_vit_environment.yml
conda activate order_vit_environment

#We install Pytorch version 1.10.0 with CUDA 11.4

Repository

/config: data and model parameter setting
/scripts: .sh file
/src: data load and augmentation, model code

How to training for Only Categorical classification and Order-learning

## Only Categorical classification
# model.name = timm model name & ../train_test: Code for validating different datasets using the best model
Using /scripts/classification.sh

## Order-learning
# ../train_test: Code for validating different datasets using the best model
Using /scripts/order_learning.sh

## feature extracture for voting (Using mamory bank)
# Feature vectors for voting through the following paths are selected in advance.: ../src/models/save_features_module.py
Using /scripts/features.sh

## voting
# sub_prob: First prob - Second prob
# trust: Meaning validated datasets described in the paper
(How to correctly predict among feature vectors extracted through features.sh and select a picture vector with a probability of 0.9 or higher at the time)
Using /scripts/voting.sh