Boundary-aware Contrastive Learning for Semi-supervised Nuclei Instance Segmentation

Installation

This code is based on mmdetection v2.18. Please install the code according to the mmdetection step first.

data preparation

BASS
├──data
|  ├──CryoNuSeg
|  |  ├──train
|  |  |  ├──mask
|  |  |  ├──patch
|  |  ├──valid
|  |  |  ├──mask
|  |  |  ├──patch
|  |  ├──test
|  |  |  ├──mask
|  |  |  ├──patch
|  |  ├──train_1_2_annotation.json
|  |  ├──train_1_4_annotation.json
|  |  ├──train_1_8_annotation.json
|  |  ├──un_train_1_2_annotation.json
|  |  ├──un_train_1_4_annotation.json
|  |  ├──un_train_1_8_annotation.json
|  |  ├──valid_1_2_annotation.json
|  |  ├──test_1_2_annotation.json

Running scripts

CryoNuseg

We take the experiment with the 1/2 labeled images for example.

First, to train the supervised model, run:

bash tools/dist_train.sh configs/noisyboundaries/cryonuseg/mask_rcnn_r50_fpn_1x_cityscapes_sup.py 1

Then, with the supervised model, generating pseudo labels for semi-supervised learning:

bash scripts/cryonuseg/extract_pl.sh 1 labels/rcity.pkl labels/cryonuseg_1_2_pl.json 

Final, perform semi-supervised learning:

bash tools/dist_train.sh configs/noisyboundaries/cryonuseg/mask_rcnn_r50_fpn_1x_coco_pl_clc.py 1