This is the TensorFlow implementation of "tGT-OCT: Toward ground-truth optical coherence tomography via three-dimensional unsupervised deep learning processing and data"
Paper doi: 10.1109/TMI.2024.3363416
In preparation; If urgent, please contact guangmingni@uestc.edu.cn
02/04/2024 Uploading the test code and optimized weights.
Install TensorFlow > 2.0.0 and other dependencies (e.g., numpy, skimage)
data:
ForTraining
: put 3D data file (.tif) for training here.ForTest
: put 3D data file for test here.ForTraining_transform
: training data after transforming will be saved here.
data_make:
data_generator.py
: performing training data transform and building tensorflow Dataset. (This file is coming soon...)
model:
- Trained model weights will be saved here.
- We have provided the optimized weights to test denoising performance on OCT retina images. Note: The provided weights were trained on volumetric data obtained by TowardPi OCT setup.
network:
ResNet3D.py:
the customized resnet with 3D convolution
output: The test output will be saved here.
train.py
: the script for running training stage. (The training code is coming soon...)
test.py
: the script for running test stage
If you want to train the tGT-OCT using our available data, you can download here: https://tianchi.aliyun.com/dataset/161472 You can select some data for test and others for training, and put them in corresponding folders.
- Running
train.py
to train a new denoising model. - Running
test.py
to test the denoising performance.
If you use this code or dataset in your research, please cite our paper.