- PyTorch implementation of Neural Style Transfer (NST) for image generation.
- The "assets" folder contains samples of input images to serve as an example to replicate the code. You need to replace the images with your own images if you want to use the proposed methodology.
- The "assets" folder also includes the saved models used for this algorithm. These files are not included in the docker image as it follows in the explanation. You need to manually replace those files locally in the same directory path as above if you choose to set up via our saved docker image.
You can set up the algorithm in either of the following two ways:
- Through Conda environment (Linux/macOS/Windows)
- Through a Docker image (Linux/macOS)
- It is recommended that you use GPU capabilities to run the codes. Nvidia/cuda should be present to be able to use the GPU.
This project is prepared with:
- Ubuntu 18.04
- CUDA 10.0 (in case of using GPU)
- Conda
To install the rest of the dependencies, you can use the conda commands:
conda env create --name=env-torch --file /NST_for_Gen/conda_env_app.yml
conda activate env-torch
Note: In case Conda cannot install the dependencies with your system, you may try to remove version specifications in the "conda_env_app.yml" file, but that may cause inconsistencies in the code.
You need to firstly request the docekr image from dockerhub, secondly build a docker container from the image, and finally run the full code.
$ sudo docker pull kenanmorani/nst_for_gen:latest
-
The docker image includes full algorithm; generation part, and training and testing segmetnation part. Thus, pulling the image may take some time.
-
To pull the docker image from dockerhub, make sure you have installed docker engine as in here for ubuntu system.
URL for the docker image is https://hub.docker.com/r/kenanmorani/nst_for_gen/tags.
-
You need to download the "assets" folder to your local directory and "cd" your local directory to run the following commands from your local directory which contain the "assets" folder.
-
To try the algorithm on your own images and masks, you need to replace the images samples in the folders with your own.
$ docker run --gpus all -it -v "$PWD/assets:/NST_for_Gen/assets" kenanmorani/nst_for_gen:latest
- If you want to try and run the code without gpu capabilities, you may detete "--gpus all" from the command above.
-
If you want to run docker as a non-root user, then you need to add your user to the docker group as in here.
-
In case of permission error for docker commands, run the command in the host terminal:
$ newgrp docker
-
For giving a user full permissions to the generated files, run the following command in the host terminal:
$ sudo chown -R
$(id -u):$ (id -g) assets/
After installation, you are ready to run the full codes to generate data, as well as training and testing on the generated data uaing our algorithm.
- This command runs on the sampled images in the 'assets' folder. Please change the directories in the code to reflect your own project directories.
$ python src/test.py --content_dir "assets/sample_input/masks/" --style_dir "assets/sample_input/data/" --style_mask_dir "assets/sample_input/masks/"For installation with Docker Image, run the container and the following command in it:
# . $bashrc >/dev/null && conda activate env-torch >/dev/null && cd /NST_for_Gen && python src/test.py --content_dir "assets/sample_input/masks/" --style_dir "assets/sample_input/data/" --style_mask_dir "assets/sample_input/masks/"- Please change the directories in the code to reflect your own project directories.
$ python src/train.py --content_dir "assets/dataset/masks" --style_dir "assets/dataset/data"For installation with Docker Image, run the container and the following command in it:
# . $bashrc >/dev/null && conda activate env-torch >/dev/null && cd /NST_for_Gen && python src/train.py --content_dir "assets/dataset/masks" --style_dir "assets/dataset/data"- Please change the directories in the code to reflect your own project directories.
$ python src/eval.py --content_dir "assets/dataset/masks" --style_dir "assets/dataset/data"For installation with Docker Image, run the container and the following command in it:
# . $bashrc >/dev/null && conda activate env-torch >/dev/null && cd /NST_for_Gen && python src/eval.py --content_dir "assets/dataset/masks" --style_dir "assets/dataset/data"- This work is supported by the Scientific and Technological Research Council of Turkey (TUBITAK) under grant no 119E578.
- The data used in this study is collected under the Marie Curie IRG grant (no: FP7 PIRG08-GA-2010-27697). It has been made publicaly available here
If you find the work useful, please consider citing the papers:
@incollection{erdem2023automated,
title={Automated analysis of phase-contrast optical microscopy time-lapse images: application to wound healing and cell motility assays of breast cancer},
author={Erdem, Yusuf Sait and Ayanzadeh, Aydin and Mayal{\i}, Berkay and Bal{\i}k{\c{c}}i, Muhammed and Belli, {"O}zge Nur and U{\c{c}}ar, Mahmut and {"O}zyusal, {"O}zden Yal{\c{c}}{\i}n and Okvur, Devrim Pesen and {"O}nal, Sevgi and Morani, Kenan and others},
booktitle={Diagnostic Biomedical Signal and Image Processing Applications with Deep Learning Methods},
pages={137--154},
year={2023},
publisher={Elsevier}
}
AND
@article{ERDEM2024106514,
title = {Novel Neural Style Transfer based data synthesis method for phase-contrast wound healing assay images},
journal = {Biomedical Signal Processing and Control},
volume = {96},
pages = {106514},
year = {2024},
issn = {1746-8094},
doi = {https://doi.org/10.1016/j.bspc.2024.106514},
url = {https://www.sciencedirect.com/science/article/pii/S174680942400572X},
author = {Yusuf Sait Erdem and Leonardo Obinna Iheme and Mahmut Uçar and Özden Yalçın Özuysal and Muhammed Balıkçı and Kenan Morani and Behçet Uğur Töreyin and Devrim Ünay},
keywords = {Neural style transfer, Generative artificial neural network, Wound healing, Phase-contrast microscopy, Biomedical image synthesis},
abstract = {Recent advancements in the field of image synthesis have led to the development of Neural Style Transfer (NST) and Generative Adversarial Networks (GANs) which have proven to be powerful tools for data augmentation and realistic data generation. While GANs have been widely used for both data augmentation and generation, NST has not been employed for data generation tasks. Nonetheless, the simpler structure of NST compared to GANs makes it a promising alternative. In this research, we introduce an NST-based method for data generation, which to the best of our knowledge, is the first of its kind. By taking advantage of simplified architecture of NST models attributed to the utilization of a real image as the style input, our method enhances performance in data generation tasks under limited resource conditions. Additionally by utilizing patch-based training and high-resolution inference process high quality images are synthesized with limited resources. Furthermore multi-model and noised input is utilized for increased diversity with the novel NST-based data generation approach. Our proposed method utilizes binary segmentation maps as the condition input, representing the cell and wound regions. We evaluate the performance of our proposed NST-based method and compare it with a modified and fine-tuned conditional GAN (C-GAN) methods for the purpose of conditional generation of phase-contrast wound healing assay images. Through a series of quantitative and qualitative analyses, we demonstrate that our NST-based method outperforms the modified C-GAN while utilizing fewer resources. Additionally, we show that our NST-based method enhances segmentation performance when used as a data augmentation method. Our findings provide compelling evidence regarding the potential of NST for data generation tasks and its superiority over traditional GAN-based methods. The NST for data generation method was implemented in Python language and will be accessible at https://github.com/IDU-CVLab/NST_for_Gen under the MIT licence.}
}