omagdy / 3drdn-cyclegan

Deep CNN for performing 3D super resolution on CT/MRI scans

3DRDN-CGAN

Introduction

This repository contains the implementation of my master thesis "Deep Learning for 3D Super-Resolution" in which a 3D CNN integrated in a Cycle-GAN architecture is used to perform super resolution on 3D models from CT/MRI scans.

Data Format Requirements

The image scans have to be in the 'nii' format and be placed in the 'data' directory where the program will recursively search through all folders there to gather all the images available.

Training

Use the command line to begin training on the dataset and select the model architecture as well as the training parameters like in the following example.

python main.py --model CYCLE-WGANGP-3DRLDSRN --learning_rate 5e-5 --batch_size 2 --lambda_adv 0.02 --epochs 100

The trained model parameters will be saved in the tensor_checkpoints directory after the training is done.

The following command can be used to check all possible input arguements and hyperparameters.

python main.py -h 

Contact

If you have any questions, please file an issue or reach me via email:

Omar Hussein: [email protected]

If my work is useful for your research, please consider citing it:

@article{article,
author = {Hussein, Omar and Sauer, Tomas},
year = {2021},
month = {07},
title = {Deep Learning for 3D Super-Resolution},
doi = {10.13140/RG.2.2.23152.58884}
}

3drdn-cyclegan's Issues

Error in valid_batch = [next(valid_dataset)]

Error:
how to solve this problem,
#Initial Random Slice Image Generation
valid_batch = [next(valid_dataset)]

Traceback (most recent call last):
File "main.py", line 26, in
main()
File "main.py", line 22, in main
LAMBDA_CYC, LAMBDA_IDT, CRIT_ITER, TRAIN_ONLY, MODEL)
File "F:\mazhiqiang\3DRDN-CycleGAN-main\training.py", line 86, in main_loop
valid_batch = [next(valid_dataset)]
File "C:\Users\user\anaconda3\envs\tensorflow-gpu2.6.0\lib\site-packages\tensorflow\python\data\ops\dataset_ops.py", line 4692, in next
return nest.map_structure(to_numpy, next(self._iterator))
File "C:\Users\user\anaconda3\envs\tensorflow-gpu2.6.0\lib\site-packages\tensorflow\python\data\ops\iterator_ops.py", line 761, in next
return self._next_internal()
File "C:\Users\user\anaconda3\envs\tensorflow-gpu2.6.0\lib\site-packages\tensorflow\python\data\ops\iterator_ops.py", line 747, in _next_internal
output_shapes=self._flat_output_shapes)
File "C:\Users\user\anaconda3\envs\tensorflow-gpu2.6.0\lib\site-packages\tensorflow\python\ops\gen_dataset_ops.py", line 2727, in iterator_get_next
_ops.raise_from_not_ok_status(e, name)
File "C:\Users\user\anaconda3\envs\tensorflow-gpu2.6.0\lib\site-packages\tensorflow\python\framework\ops.py", line 6941, in raise_from_not_ok_status
six.raise_from(core._status_to_exception(e.code, message), None)
File "", line 3, in raise_from
tensorflow.python.framework.errors_impl.InvalidArgumentError: Need minval < maxval, got 50 >= 14
[[{{node StatefulPartitionedCall/StatefulPartitionedCall/random_uniform_2}}]] [Op:IteratorGetNext]

Inference

Could you pleaes help me with the inference problem, I have no idea of how to inference the model which I have trained well. I could not find the ckpt.meta file.

Thank you !

Can you provide pretrained weights?

My hardware is insufficient for training and I want to try inference before I consider using it for research.
Is it possible to share your model weights or provide a pretrained model?

How to obtain the data set?

if I just want to generate thin slice CTs，how can I do?

Just remove the SCALING_FACTOR for x and y, and keep the SCALING_FACTOR for z, Will it work if I just want to generate thin slice CTs?
lr_image = tf.image.resize(blurred_image, [x//SCALING_FACTOR, y//SCALING_FACTOR], method=interpolation_method).numpy()
lr_image = np.rot90(lr_image, axes=(1,2))
lr_image = tf.image.resize(lr_image, [x//SCALING_FACTOR, z//SCALING_FACTOR], method=interpolation_method).numpy()
ups_lr_image = tf.image.resize(lr_image, [x//SCALING_FACTOR, z], method=interpolation_method).numpy()
ups_lr_image = np.rot90(ups_lr_image, axes=(1,2))