Generating-disaster-image-from-input-image-using-DiscoGAN
Using DiscoGAN to generate image of how houses and landscapes would look like after a natural disaster make people realize the personal cost they might have to pay due to climate change
Prerequisites
- Python 3.7.10
- PyTorch
- Numpy/Scipy/Pandas
- Progressbar
- OpenCV
Loading dataset
Run the following code snippet in the notebook
import os
!mkdir './dataset/houses'
!mkdir './dataset/disasters'
Download, unzip and store the data as follows:
https://www.kaggle.com/amir22010/house-price
in "houses" directory
https://www.kaggle.com/mikolajbabula/disaster-images-dataset-cnn-model
Only the earthquake images in "disasters" directory
Training DiscoGAN
Run the cells in the notebook (./discogan/model notebook) in order.
Results
The training data is described above and we used random images of houses from Google for testing. Some of the generated images are shown above.
Purpose
Climate change will eventually affect all 7 billion of us on Earth. Yet, only a few people realize the urgency to act. This project is intended to help people visualize how the places they love might look if climate change continues. It is said, "a picture is worth a thousand words", so, these generated pictures might be able to work the magic on some minds that a world wide web full of statistics could not. :)