particle-ViT-segmentation

Challenge resolution by appling segmentation techniques.

Directories layout:

particle-ViT-segmentation
├── Config
│   └── ...
├── Dataset
│   └── install_dataset.py
├── preprocessing
│   ├── analyser.py
│   ├── creation.py
│   ├── Dataset.py
│   └── segmentation.py
├── techniques
│   ├── unet
│   └── vit
├── ut
│   └── ...
├── utils
│   └── ...
├── main.py
├── create_dataset.py
├── requirements.txt
└── README.md

Installation

Make sure you have all the necessary requirements to use this repo.

git clone https://github.com/AngeloDamante/particle-ViT-segmentation.git
cd particle-ViT-segmentation
pip3 install -r requirements.txt

Dataset

The starting datasets for challenge can be found here. For your convenience, it can be easily downloaded.

# in particle-ViT-segmentation directory
python3 Dataset/install_dataset.py

Preprocess

The dataset was generated by performing convolution between a black image with white points centered in the groundtruth values and a Gaussian filter. To generate raw data, we have written create_dataset.py script that you can use with input json file for config settings.

usage: create_dataset.py [-h] [-C CONFIG]

optional arguments:
  -h, --help            show this help message and exit
  -C CONFIG, --config CONFIG name of config in CONFIG_DIR

The raw data generated have the structure below:

[
    img:np.ndarray,
    target:np.ndarray,
    gth:List[Particle],
    snr:SNR,
    density:Density,
    t:int
]

UNET

TODO

ViT

TODO