Revisiting 3D Context Modeling with Supervised Pre-training for Universal Lesion Detection on CT Slices
This is an implementation of MICCAI 2020 paper Revisiting 3D Context Modeling with Supervised Pre-training for Universal Lesion Detection on CT Slices.
This code is based on MMDetection. Please see it for installation.
Download DeepLesion dataset here.
We provide coco-style json annotation files converted from DeepLesion. Please download json files here, unzip Images_png.zip and make sure to put files as following sturcture:
data
├──DeepLesion
├── annotations
│ ├── deeplesion_train.json
│ ├── deeplesion_test.json
│ ├── deeplesion_val.json
└── Images_png
└── Images_png
│ ├── 000001_01_01
│ ├── 000001_03_01
│ ├── ...
We provide models pre-trained on COCO dataset which can be used for different 3D medical image detection.
The pre-trained MP3D63 model can be downloaded from BaiduYun(verification code: bbrc) or GoogleDrive.
To train MP3D & P3d model on deeplesion dataset, run:
bash tools/dist_train.sh configs/deeplesion/mp3d_groupconv.py 8
bash tools/dist_train.sh configs/deeplesion/p3d.py 8
If you have questions or suggestions, please open an issue here.