AI4Good HS21 - Group 3b - Rainforest deforestation monitoring

Setup

Create virtual environment and install the dependencies using pip:

pip install -r requirements.txt

Before running any of the scripts, add the root of this repository to your PYTHONPATH environment variable.

Create the dataset

To create the deforestation training dataset mentioned in the report, run the following script:

python3 data/create_deforestation_dataset.py \
    --name <DATASET_NAME> \
    -sd 2020-08-01 \
    -ed 2020-12-31 \
    --log-scale \
    -aoi sampling_rectangles_2 \
    --scale 10 \
    --num-samples-per-area 400 \
    --months-before-acquisition 8

or download it directly from here https://polybox.ethz.ch/index.php/s/pcQ6TSYS5nKJx9q.

Training

To train the deforestation detection model:

python3 train.py \
    -d <PATH_TO_DATASET_DIRECTORY> \
    -o <PATH_TO_CHECKPOINTS_FOLDER> \
    --backbone efficientnet-b3 \
    -m deeplabv3+