Extracting Quantitative Streamline Information from Surface Flow Visualization Images in a Linear Cascade using Convolutional Neural Networks.

These files contain the code for Paper Extracting Quantitative Streamline Information from Surface Flow Visualization Images in a Linear Cascade using Convolutional Neural Networks.

I developed it on Linux, so it might be a bit problematic on Windows. Don't try to train without a GPU

The files are provided in the hope that they will be useful to other researchers but we provide the software ``AS IS'' WITHOUT WARRANTY OF ANY KIND. The code is meant to be a demonstration of the techniques used in the paper. I think my idea might be good but my coding skill is really terrible. Thus, do some modification according you task, and hopefully it might work.

Requirements

• Python 3.7
• Pytorch >=1.4 (Last test 1.9)
• OpenCV
• Matplotlib
• Kornia
• Other package like Numpy, h5py, PIL, json, skimage.

Project Architecture

Camera Lens Calibration -> contains the calibration images and the python code for the lens correction and coordinate detection.
train.py -> train the egde detection neural networks
test.py -> test the egde detection neural networks
preprocessing.py -> including calibrated visualization image, division into cells, grayscale conversion, image smoothing, and exposing intensity gradients. 
postprocessing.py -> get your flow direction result

Before to start please check dataset.py, from the first line of code you can see the datasets used for training/testing. Before you generate your own result, please check again.

Datasets used for Training

Edge detection datasets

• BIPED

How to use

1. After completing the setup, use preprocessing.py to preprocess the images to be processed.
2. Download the dataset and check the Parser settings, then run train.py.
3. Choose the appropriate stride and neighborhood size (they need to match the size of the images), then run postprocessing.py.

Tips: If you do not want to train a neural network from scratch, or if your device does not have sufficient performance, you can also use the results of Canny edge detection to run postprocessing.py.

Acknowledgement

• We like to thanks to the previous repo: LDC

Citation

If you like this work please cite these papers if you find helpful in your academic/scientific publication,

@ARTICLE{xsoria2022ldc,
  author={Soria, Xavier and Pomboza-Junez, Gonzalo and Sappa, Angel Domingo},
  journal={IEEE Access}, 
  title={LDC: Lightweight Dense CNN for Edge Detection}, 
  year={2022},
  volume={10},
  number={},
  pages={68281-68290},
  doi={10.1109/ACCESS.2022.3186344}}

@article{abdelsalam2017exploiting,
  title={Exploiting Modern Image Processing In Surface Flow Visualisation},
  author={Abdelsalam, Tarek Ihab and Williams, Richard and Ingram, Grant},
  year={2017}
}