An official implementation of paper "Semi-Supervised Cell Recognition under Point Supervision"

Setup

Python 3.7

pip install -r requirements.txt 

Data preparation

Two choices.

• You can download the raw data from CoNIC to datasets/conic folder and then run this script to obtain training/validation/test subsets .
• A more convenient way is to download the ready-made data subsets from Google Drive (after review).

Train

To reproduce baseline models:

python train_base.py --dataset conic --space 8 --num_classes 6 --eos_coef 0.4 --match_dis 6 --output_dir=he_sup_5_base --ratio 5

To train PCR models under our proposed framework:

python train_semi.py --dataset conic --space 8 --num_classes 6 --eos_coef 0.4 --match_dis 6 --output_dir=he_sup_5_semi --ratio 5 --enable_semi_sup

Test

To test baseline models, run

python train_base.py --dataset conic --space 8 --num_classes 6 --match_dis 6 --ratio 5 --test

To test models trained using our framework, run

python train_semi.py --dataset conic --space 8 --num_classes 6 --match_dis 6 --ratio 5 --test

Checkpoints

The checkpoints will be also released here after review.