This project contains work of point-supervised segmentation of microscopy images and volumes via objectness regularization and re-implementations of PseudoEdgeNet(paper) and LCFCN(paper|code). The LCFCN part contains most of the original code while the PseudoEdgeNet was implemented according to the paper as we weren't able to find existing code from the project.
To pre-setup the environment, we recommand to use
conda env create -f environment.yml
and
pip install -r requirements.txt
This command installs pydicom and the Haven library which helps in managing the experiments.
A stain norm process need to be done in the preprocessing step before generating objectness map.
git clone https://github.com/wanghao14/Stain_Normalization.git
You can download the BBBC, TNBC or MoNuSegTraingData to the directory. It contains the splitted index for each validation fold and the dataset.
Before training, the objectness map for each image should be generated first. Make sure the images ,point annotations and the groundtruth (for validation) saved in the path as below:
Dataset
└── Train
└── Images
└── Points
└── GroundTruth
└── Validate
└── ...
└── Test
└── ...
Then run generate_obj.py to generate objectness maps.
Train and validate the model with BBBC dataset
python trainval.py -d BBBC -e exp_config_ponet.json -r 1
Train and validate the model with TNBC dataset
python trainval.py -d TNBC -e exp_config_ponet.json -r 1
Train and validate the pseudoedgenet model with MoNuSeg dataset
python trainval.py -d MoNuSegTrainingData -e exp_config_ponet.json -r 1
Both trainval.py and trainval.ipynb can be used to train and validate the model. In trainval.ipynb, you can get a brief introduction of the setup of the dataset.
The models are defined in src/models. The dataset is defined in src/datasets/__init__.py. The loss function is defined in lcfcn/lcfcn_loss.py. The function compute_weighted_crossentropy is used in the PseudoEdgeNet model. The compute_lcfcn_loss is used in LCFCN model.