m1dsolo / UBC-OCEAN-7th

UBC-OCEAN 7th place solution

UBC Ovarian Cancer Subtype Classification and Outlier Detection (UBC-OCEAN) 7th place solution.

This code repository mainly includes data preprocessing and multiple instance learning(MIL) model training.

• For detailed competition information, please refer to the competition link.
• For detailed solution information, please refer to the discussion link.
• For detailed submission notebook, please refer to the notebook link.

Requirements

python>=3.10

Install

git clone --recurse-submodules https://github.com/m1dsolo/UBC-OCEAN-7th.git
pip install UBC-OCEAN-7th/yangdl

Pipeline

1. Tiling patches

   python png2patches.py

2. Download ctranspath checkpoints

3. Extract features

   python gen_features.py -t ctrans
   python gen_features.py -t vits16

4. train MIL

   python train_dsmil.py -t ctrans
   python train_dsmil.py -t vits16
   python train_perceiver.py -t ctrans
   python train_perceiver.py -t vits16

model will save to ./res/*/ckpt

For more information please refer to:

1. CTransPath, MIA2022
2. LunitDINO, CVPR2023
3. DSMIL, CVPR2021
4. Perceiver, BMVA2023