Skip to content

Atzetra/UC-Project

Repository files navigation

logo

Implementing Image Segmentation into AutoKeras


Installation

To install the dependencies, run

conda create --file autokeras.yml
pip install -e .

This repo has been developed in a WSL/Ubuntu environment using this dockerimage for cuda compute compatabilities.

Structure

All the files needed for dataset generation of the water dataset are located in ./helpers/datagen.py. This file assumes that you have the Sentinel 2 dataset downloaded and preprocessed.

Use the ./datasets/dataset_processing.ipynb notebook for this. Here we detailed how to download and generate the files needed.

To run the sementation, run ./segmentation.py. I (@Atzetra) got to the point we had to define output shapes and had to choose our own segmentation blocks/architecture to incorporate as detailed in our report. A stacktrace when running /segmentation.py should show you what we mean.

Running ./segmentation.py should be enough to run if all the paths are correct. This script generates the tfds pipeline and preprocesses all data if the steps above for generating the sentinel-2 dataset are followed.

Acknowledgements

I would like to thank the Urban Computing team for offering a very interesting course to be able to follow. As someone of the master's Bio-Pharmaceutical Sciences, it was fun, albeit very challanging, to learn.

  • Enzo

About

Urban Computing Project

Resources

License

Stars

Watchers

Forks

Contributors