Train and compare YOLO segmentation variants (Ultralytics) for different datasets and augmentation settings. This project was built specifically for the ARLab at the University of Augsburg and developed in the context of the Zirbi robot.
Training data is shipped as a single archive in the repository root: datasets.zip. After cloning, extract it here (repo root) so the dataset folders (e.g. data_640/, each with a data.yaml) sit alongside scripts/ and docs/. Details: docs/startup-local.md.
The intended workflow for most users is local training via scripts/local/ (see docs/startup-local.md). SLURM scripts under scripts/slurm/ are optional: they are provided for running jobs on university compute clusters (e.g. LICCA at the University of Augsburg) and are documented in docs/startup-slurm.md.
- Local setup / training (primary):
docs/startup-local.md - SLURM / cluster jobs (optional):
docs/startup-slurm.md
Python entry points:
scripts/local/train_models.py
SLURM job scripts (cluster only):
scripts/slurm/slurm_train_*.sh,scripts/slurm/slurm_compare_*.sh
- Meruna Yugarajah m.yugarajah@gmail.com
- Aleksander Michalak aleksander1.michalak@uni-a.de