Shelf-Bench dataset repository 🇦🇶

Dataset Information

Dataset provided by: C. Baumhoer, DLR
Download the Shelf-Bench dataset from: https://zenodo.org/records/17610871

Link to the original paper:

Our trained models are stored in the trained_models.zip

1. Create a virtual environment and install requirements.txt into your virtual environment

With python

python3 -m venv .venv
source .venv/bin/activate

Or with uv

uv venv
source .venv/bin/activate

pip install -r requirements.txt

2. Preprocess all images and masks

python data_preprocessing.py

or

uv data_preprocessing.py

3. Apply trained_models to the Shelf-Bench dataset, or train from scratch using:

python continuous_train.py -m model.name=CHOOSE MODELS training.loss_function=CHOOSE LOSS FUNCTIONS

or

uv run continuous_train.py -m model.name=CHOOSE MODELS training.loss_function=CHOOSE LOSS FUNCTIONS

This code uses Wandb which can be set up using: https://wandb.ai/site/