Gear8 Dataset

Gear8 is an industrial object detection dataset for 8 types of components across two models of Bosch power window motor gear reducers.

This dataset is featured in the paper MICA: Multi-Agent Industrial Coordination Assistant and was originally proposed in Snap, Segment, Deploy: A Visual Data and Detection Pipeline for Wearable Industrial Assistants (SMC 2025).

GitHub Repository: https://github.com/Kratos-Wen/MICA

---

Data Structure & Extraction

The training images are provided in 4 sharded archives due to their large size (~52GB). Follow these steps to merge and extract the full dataset.

1. File List

• train_part_1
• train_part_2
• train_part_3
• train_part_4
• val.tar
• annotations.tar

2. Extraction Commands

Run the following in your terminal:

# Merge and extract training images
cat train_part_1 train_part_2 train_part_3 train_part_4 > train_images.tar
tar -xvf train_images.tar

# Extract validation set and annotations
tar -xvf val.tar
tar -xvf annotations.tar

3. Directory Layout

After extraction, ensure your directory is structured as follows:

Gear8/
├── gear8.yaml           # Configuration file
├── images/
│   ├── train/         # Training images (from merged shards)
│   └── val/         # Validation images (from val.tar)
└── labels/
    ├── train/        # Training labels (from annotations.tar)  
    └── val/         # Validation labels (from val.tar)

---

Citation

If you use this dataset, please cite the following works:

@article{wen2025mica,
  title={MICA: Multi-Agent Industrial Coordination Assistant},
  author={Wen, Di and Peng, Kunyu and Zheng, Junwei and Chen, Yufan and Shi, Yitian and Wei, Jiale and Liu, Ruiping and Yang, Kailun and Stiefelhagen, Rainer},
  journal={arXiv preprint arXiv:2509.15237},
  year={2025},
  doi={10.48550/arXiv.2509.15237}
}

@inproceedings{wen2025snap,
  title={Snap, Segment, Deploy: A Visual Data and Detection Pipeline for Wearable Industrial Assistants},
  author={Wen, Di and Zheng, Junwei and Liu, Ruiping and Xu, Yi and Peng, Kunyu and Stiefelhagen, Rainer},
  booktitle={2025 IEEE International Conference on Systems, Man, and Cybernetics (SMC)},
  pages={1270--1276},
  year={2025},
  organization={IEEE}
}