Together with the data, we provide also a development kit (devkit) including a PyTorch dataloader, scripts for computing the evaluation metrics, and scripts for visualizing predictions. See https://github.com/PRBonn/phenobench.

More specifically, we provide the following code in the devkit:

• PyTorch-based dataloader that loads the data from the directories and converts the data into corresponding tensor representation for training.
• Evaluation script for all benchmarks to compute the metrics in a reproducible way. Furthermore, these evaluations scripts will also be used for the server-side evaluation, which ensures that we have consistent numbers.
• Scripts for visualization to render results, but also visualize the original data.

For reproducibility, we provide also the code of the baselines or instructions to run the baselines including the configuration files for training the baselines. See https://github.com/PRBonn/phenobench-baselines.

For each of the baselines, we provide instructions or modified code that enables to train the models. Please refer to the corresponding README.md in each subdirectory for instructions to run the baselines.

Additionally, we also provide the checkpoints of the trained baselines models and the predictions of each baseline on the validation and test set.