Blind Denoising of Fluorescence Microscopy Images Using GAN-Based Global Noise Modeling

This repos provides a denoising method for fluorescence microscopy images based on GAN global noise modeling.

Framework of our denoising method

Dependencies

• python 3
• torch>=1.4.0
• torchvision>=0.5.0
• dominate>=2.4.0
• visdom>=0.1.8.8
• matlab

How to use

1. P2P-NM train/test

• Put source data into folder P2P-NM/datasets/ (e.g. our dataset named mask_mito_1080)
• To view training results and loss plots, run python -m visdom.server
• Train a model

python train.py --batch_size 1 --model pix2pix --direction BtoA --dataroot ./datasets/ --phase mask_mito_1080 --name trained_on_1080 --niter 500 --niter_decay 500   

More options are list in options/base_options.py and train_options.py
Model weights will be saved at P2P-NM/checkpoints/trained_on_1080

• Run Data_process_matlab/make_random_mask_s1.m to generate test dataset named edge_random_masks
• Test the model using edge_random_masks

python test.py --model pix2pix --direction BtoA --num_test 99999 --dataroot ./datasets/ --phase edge_random_masks --name trained_on_1080

More options are list in options/base_options.py and test_options.py

• The test results will be saved at P2P-NM/results/trained_on_1080/edge_random_masks_latest

2. Build data for training P2P-DN

• Put P2P-NM results edge_random_masks_latest/ into Data_process_matlab/
• Run get_global_noise_s2.m to output global_noise
• Prepare 1080 masks
• Run make_denoise_training_s3.m to build dataset denoise_train_1080 to train P2P-DN

3. P2P-DN train/test

• Put dataset denoise_train_1080 into P2P-DN/datasets/
• Train a model

python train.py --batch_size 1 --model pix2pix --direction BtoA --dataroot ./datasets/ --phase denoise_train_1080 --name trained_on_1080 --niter 100 --niter_decay 100

Weights will be saved at P2P-DN/checkpoints/trained_on_1080

• Test the model using real fluorescence microscopy images

python test.py --model pix2pix --direction BtoA --num_test 99999 --dataroot ./datasets/ --phase mito_real_156 --name trained_on_1080

• Results will be saved at P2P-DN/results/trained_on_1080/

Contributing

Code for this projects developped at CBMI Group (Computational Biology and Machine Intelligence Group).
CBMI at National Laboratory of Pattern Recognition, INSTITUTE OF AUTOMATION, CHINESE ACADEMY OF SCIENCES.
Bug reports and pull requests are welcome on GitHub at https://github.com/cbmi-group/BlindDenoising