Weed Mapping in Multispectral Drone Imagery using Lightweight Vision Transformers

Github repo for the paper "Weed Mapping in Multispectral Drone Imagery using Lightweight Vision Transformers"

Installation

Recommended to create a Python virtual environment. Install the requirements

pip install -r requirements.txt

Experiment reproduction

Download the WeedMap dataset

py wd.py download

Or directly download the tiles from here

Directory tree should be like this:

dataset/
├─ raw/
│  ├─ RedEdge/
│  ├─ Sequoia/

Preprocessing

py wd.py preprocess --subset RedEdge
py wd.py preprocess --subset Sequoia

Data augmentation

py wd.py augment --subset RedEdge
py wd.py augment --subset Sequoia

Training and Testing

python wd.py experiment --file params/Rheinbach/SplitLawin.yaml

Refer to the params folder for the different experiments made