CircleNet: Anchor-free Detection with Circle Representation

The official implementation of CircleNet, MICCAI 2020

[PyTorch] [project page] [MICCAI paper]

Object detection networks are powerful in computer vision, but not necessarily optimized for biomedical object detection. In this work, we propose CircleNet, a simple anchor-free detection method with circle representation for detection of the ball-shaped glomerulus. Different from the traditional bounding box based detection method, the bounding circle (1) reduces the degrees of freedom of detection representation, (2) is naturally rotation invariant, (3) and optimized for ball-shaped objects.

full citation is

Haichun Yang, Ruining Deng, Yuzhe Lu, Zheyu Zhu, Ye Chen, Joseph T. Roland, Le Lu, Bennett A. Landman, Agnes B. Fogo, and Yuankai Huo. "CircleNet: Anchor-free Detection with Circle Representation." arXiv preprint arXiv:2006.02474 (2020).

Envrioment Set up

We used CUDA 10.2 and PyTorch 0.4.1.

The implementation is based on the CenterNet. https://github.com/xingyizhou/CenterNet

Please install the packages, following https://github.com/xingyizhou/CenterNet/blob/master/readme/INSTALL.md

• For "Clone this repo" step, please clone CircleNet rather than CenterNet

Testing on a whole slide image

The Case 03-1.scn file is avilable https://vanderbilt.box.com/s/s530m45rvk626xi1thwcdc2bhoea758r

The model_10.pth model file is avilable (human kidney) https://vumc.box.com/s/wpar2kz9600h9ao3wowjzc3y50znneop

To run it on a testing scan, please go to "src" folder and run

python run_detection_for_scn.py circledet --arch dla_34 --demo "/media/huoy1/48EAE4F7EAE4E264/Projects/from_haichun/batch_1_data/scn/Case 03-1.scn" --load_model /media/huoy1/48EAE4F7EAE4E264/Projects/detection/CircleNet/exp/circledet/kidpath_dla_batch4/model_10.pth --filter_boarder --demo_dir "/media/huoy1/48EAE4F7EAE4E264/Projects/detection/test_demo"

The demo_dir is output dir, which you set anywhere in your computer.

After running code, you will see a Case 03-1.xml file. Then you put the xml and scn files into the same folder, and open the scn file using ImageScope software (only avilable in Windows OS), you can see something like the following image, with green detection results.

A Google Colab demo of the above testing code is added

https://github.com/hrlblab/CircleNet/blob/master/src/circle_net_demo.ipynb

Run your own training code

The training code is

python main.py circledet --exp_id kidpath_dla_batch4 --arch dla_34 --batch_size 4 --master_batch 4 --lr 2.5e-4   --gpus 0 --print_iter 1  --dataset kidpath --save_all --load_model ../models/ctdet_coco_dla_2x.pth

You can get the ctdet_coco_dla_2x.pth model from model zoo https://github.com/xingyizhou/CenterNet/blob/master/readme/MODEL_ZOO.md