VIEW-DISENTANGLED TRANSFORMER FOR BRAIN LESION DETECTION-ISBI 2022

This repo contains the supported code and configuration files of View-Disentangled Transformer. It is based on Swin-Transformer-Object-Detection.

Pretrain Model

Model	config	Params
Baseline	config	google drive
+VDFormer	config	google drive

Prerequisites

Linux
Python 3.7.11
Pytorch 1.9.1
CUDA 10.2
MMDetection 2.11.0
MMCV 1.3.14

Installation

Prepare environment

Create a conda virtual environment and activate it.

conda create -n VDFormer python=3.7 -y
conda activate VDFormer

Install PyTorch and torchvision following the official instructions, e.g.,
```
conda install pytorch cudatoolkit=10.2 torchvision -c pytorch
```

Install MMDetection

Install mmcv-full

pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/{cu102}/{torch1.9.0}/index.html

Install MMDetection-VDFormer

git clone $https://github.com/VDFormer(The github link of VD Former)
cd $VDFormer(The folder of VD Former)
pip install -r requirements/build.txt
python setup.py develop

Dataset

We use dataset in coco format.

If you want to train VD-Former on other data, you need to change the format to coco and the data path in config/VDFormer/base_config.py

Model definition

The VDFormer model is mainly defined in mmdet/models/swin_transformer_proposed.py and mmdet/models/necks/swin_fusion_layer_proposed.py

Inference

# sinlge-gpu testing
python tools/test.py <CONFIG_FILE> <DET_CHECKPOINT_FILE> --eval bbox

Training

# sinlge-gpu training
python tools/train.py <CONFIG_FILE> --cfg-options model.pretrained=<PRETRAIN_MODEL>
# multi-gpu training
CUDA_VISIBLE_DEVICES=0,1 tools/dist_train.sh <CONFIG_FILE> 2 --cfg-options model.pretrained=<PRETRAIN_MODEL>

Citing VD Transformer

@artical{li2022VDFormer,
    title={VIEW-DISENTANGLED TRANSFORMER FOR BRAIN LESION DETECTION},
    year={2022}
}

Contact

2022.3.12 
Junjia Huang
1959643995@qq.com

lhaof/ISBI-VDFormer