DeepdespeckeNet

Workflow overview

Network structure

System requirements

Before training or testing, you need to prepare the deep-learning environment:

Python 3.9
CPU or NVIDIA GPU + CUDA CuDNN
Linux OS

Installation guide

You should prepare at least 2 2080ti gpus or change the batch size in training process.

pip install -r requirement.txt

Installation will be completed quickly

Demo

You can download part of the datasets on Google Drive, and we will open souce the entire data soon.

For testing dataset, please put it on "../datasets/test/xxx"

For training dataset, please put it on "../datasets/train/xxx"

Due to the upload file size limitation, we provide representative despeckle image pairs and weight files over the c666-1 dataset in the zip file. Unzip it and run:

python3 test.py --dataroot nasopharyngeal_carcinoma_cell  --checkpoints_dir ./checkpoints/ckp_C666_complex  --ngf 64 --results_dir ./results/ckp_C666_complex

Running time is about 20s when using a single 2080Ti graphics card. The default results are saved in "./results/xxx/dif2IP/test_latest/images", where xxx_fake_xxx are the outputs of our network, xxx_real_xxx are the ground truths, and the xxx_real_A.bmp is the speckle image. You can also open "./results/xxx/dif2IP/test_latest/images/index.html" to check the results.

Instruction for use

Training process

Run

python3 train.py --dataroot xxx --gpu_ids 0,1  --checkpoints_dir ./checkpoints/xxx --batch_size 16  --ngf 64  --no_flip

the traing dataroot should be one of the [breast_cancer_cell, hela_cell, breast_cancer_tissue, nasopharyngeal_carcinoma_cell, simu_scatter4,simu_scatter6,simu_scatter8, simu_scatter10, simu_scatter12]

You can download the pre-trained weights files on Google Drive. Put the weight file on "./checkpoints/".

Testing process

Run:

python3 test.py --dataroot xxx  --checkpoints_dir ./checkpoints/xxx  --ngf 64 --results_dir ./results/xxx

The testing dataroot should be one of the [breast_cancer_cell, hela_cell, breast_cancer_tissue, breast_cancer_tissue_2048, nasopharyngeal_carcinoma_cell, simu_scatter4,simu_scatter6,simu_scatter8, simu_scatter10, simu_scatter12]

Common problem

When you encounter any problems please contact us by email [email protected] or [email protected]