solafune_deforestation_baseline

This repository provides a baseline solution for the Identifying Deforestation Drivers competition by Solafune.

By running the provided notebooks, you will achieve a score of around 0.533 on the public leaderboard.

cf. @solafune (https://solafune.com) Use for any purpose other than participation in the competition or commercial use is prohibited. If you would like to use them for any of the above purposes, please contact us.

Requirements

Datasets

Download the datasets and organize them as follows:

data/
├── evaluation_images/
│   ├── evaluation_0.tif
│   ├── evaluation_1.tif
│   ├── evaluation_2.tif
│   ├── ...
├── train_images/
│   ├── train_0.tif
│   ├── train_1.tif
│   ├── train_2.tif
│   ├── ...
├── train_annotations.json

Libraries

Please install the python packages imported the first cell of the notebooks.

Usage

Step-1: generate segmentation masks for training

generate_masks.ipynb generates segmentation masks for the training images and save them as .npy files. These files will be used as training labels (see the section below).

Each .npy file contains a numpy array of shape (4, 1024, 1024) for the four classes (grassland_shrubland, logging, mining, and plantation). The pixels with value 255 are considered to be the corresponding class.

data/
├── train_masks/
│   ├── train_0.npy
│   ├── train_1.npy
│   ├── train_2.npy
│   ├── ...

The notebook also saves the visualization of the masks along with the RGB image as a png file. These files are just for visualization (not used for training).

data/
├── vis_train/
│   ├── train_0.png
│   ├── train_1.png
│   ├── train_2.png
│   ├── ...

Below is a visualization example for train_39.tif.

Step-2: train U-Net and generate the submission file for the evaluation images

train_model.ipynb trains a U-Net model for 4-class segmentation (grassland_shrubland, logging, mining, and plantation) and generates a submission JSON file for the evaluation images from the output from the U-Net model.

The submission JSON file is saved to data/submission.json.

Before running this notebook, you have to run generate_masks.ipynb to generate .npy files which will be used for training.