HibikinoMusashiHome/hma_rc26_object_images_dataset
HMA RC26 Incheon RGB Image Dataset Dataset Overview This dataset combines synthetic images and annotated RGB images captured with ASUS Xtion Pro Live and ORBBEC Gemini 336L. It was used to train a YOLOv8 segmentation model for object recognition at RoboCup@Home 2026 in Incheon (RC26) [https://github.com/RoboCupAtHome/Incheon2026]. Data Collection Synthetic Data The simulator generated 1,000 images using domain randomization. Each… See the full description on the dataset page: https://huggingface.co/datasets/HibikinoMusashiHome/hma_rc26_object_images_dataset.
HMA RC26 Incheon RGB Image Dataset
<!-- TODO: データセットの概要、収集方法、ファイル構成、クラス一覧などを記載 -->
Dataset Overview
This dataset combines synthetic images and annotated RGB images captured with ASUS Xtion Pro Live and ORBBEC Gemini 336L. It was used to train a YOLOv8 segmentation model for object recognition at RoboCup@Home 2026 in Incheon (RC26) [https://github.com/RoboCupAtHome/Incheon2026].
Data Collection
Synthetic Data
The simulator generated 1,000 images using domain randomization. Each image has COCO JSON and YOLO segmentation TXT annotations. The copies under the Xtion and Gemini directories are identical.
Real-World Data
Real RGB images were captured with ASUS Xtion and Gemini cameras, manually annotated in LabelMe, and converted to YOLO segmentation format.
Training used one copy of the synthetic data together with the real images from both cameras.
File Structure
.
|-- README.md
|-- rc26_dataset_xtion/
| |-- generate_data/
| | |-- data/
| | | |-- 0/
| | | | `-- 0..99/
| | | |-- 1/
| | | | `-- 0..99/
| | | |-- 2/
| | | | `-- 0..99/
| | | |-- 3/
| | | | `-- 0..99/
| | | |-- 4/
| | | | `-- 0..99/
| | | |-- 5/
| | | | `-- 0..99/
| | | |-- 6/
| | | | `-- 0..99/
| | | |-- 7/
| | | | `-- 0..99/
| | | |-- 8/
| | | | `-- 0..99/
| | | `-- 9/
| | | `-- 0..99/
| | |-- annotations_1000.json
| | |-- annotations_1000.txt
| | |-- annotations_template.json
| | |-- class_names.yaml
| | `-- main.py
| `-- real_images/
| |-- images/
| | |-- train/
| | |-- val/
| | `-- test/
| |-- labels/
| | |-- train/
| | |-- val/
| | `-- test/
| |-- raw/
| `-- split_meta.yaml
`-- rc26_dataset_gemini/
|-- generate_data/
| |-- data/
| | |-- 0/
| | | `-- 0..99/
| | |-- 1/
| | | `-- 0..99/
| | |-- 2/
| | | `-- 0..99/
| | |-- 3/
| | | `-- 0..99/
| | |-- 4/
| | | `-- 0..99/
| | |-- 5/
| | | `-- 0..99/
| | |-- 6/
| | | `-- 0..99/
| | |-- 7/
| | | `-- 0..99/
| | |-- 8/
| | | `-- 0..99/
| | `-- 9/
| | `-- 0..99/
| |-- annotations_1000.json
| |-- annotations_1000.txt
| |-- annotations_template.json
| |-- class_names.yaml
| `-- main.py
`-- real_images/
|-- images/
| |-- train/
| |-- val/
| `-- test/
|-- labels/
| |-- train/
| |-- val/
| `-- test/
|-- raw/
`-- split_meta.yamlEach 0..99/ entry represents 100 sample directories. Every sample contains a JPG image and matching JSON and TXT annotations.
File Descriptions
generate_data/data: synthetic images and per-image annotations.annotations_1000.json: combined COCO annotations.annotations_1000.txt: synthetic-image path list.annotations_template.json: COCO category template.class_names.yaml: ordered YOLO classes.main.py: simulator and domain-randomization settings.real_images/imagesandlabels: split real images and YOLO labels.real_images/raw: original images and LabelMe JSON files.split_meta.yaml: split settings and file assignments.
Class List
Rows are ordered by zero-based class ID (0–25).
<!--https://github.com/RoboCupAtHome/Incheon2026/blob/main/objects/objects.mdに準拠している--> Its 26 classes were selected from the official RC26 object set.
License
This dataset is released under CC BY-NC 4.0.
You may use, modify, and redistribute it for non-commercial purposes (academic research, education, and non-profit use) as long as you provide attribution by citing the works listed below.
Commercial use is not permitted without prior written permission. For commercial licensing inquiries, please contact hma@brain.kyutech.ac.jp.
Citation
Attribution is satisfied by citing the following works:
@inproceedings{maruno2026,
author = {Ryouga Maruno and Akihiro Suzuki and Naoki Yamaguchi and Tomoya Shiba and Hakaru Tamukoh},
title = {Automatic Skeleton Definition Framework for Flexible and Amorphous Objects},
booktitle = {Proceedings of the 18th International Conference on Computer and Automation Engineering (ICCAE2026)},
year = {2026}
}
@inproceedings{shiba2025,
author = {Tomoya Shiba and Naoki Yamaguchi and Ryoga Maruno and Akihiro Suzuki and Hakaru Tamukoh},
title = {Pose Diversity Improves Object Detection Generalization},
booktitle = {Proceedings of the 2025 International Conference on Artificial Life and Robotics (ICAROB2025)},
year = {2025},
pages = {245--250}
}
@article{ono2022,
author = {Tomohiro Ono and Daiju Kanaoka and Tomoya Shiba and Shoshi Tokuno and Yuga Yano and Akinobu Mizutani and Ichiro Matsumoto and Hayato Amano and Hakaru Tamukoh},
title = {Solution of World Robot Challenge 2020 Partner Robot Challenge (Real Space)},
journal = {Advanced Robotics},
volume = {36},
number = {17--18},
pages = {870--889},
year = {2022},
doi = {10.1080/01691864.2022.2115315}
}
@article{kobayashi2025,
author = {Ryohei Kobayashi and Kosei Isomoto and Kosei Yamao and Soma Fumoto and Koshun Arimura and Naoki Yamaguchi and Akinobu Mizutani and Tomoya Shiba and Kouki Kimizuka and Yuta Ohno and Ryo Terashima and Hiromasa Yamaguchi and Tomoaki Fujino and Ryoga Maruno and Wataru Yoshimura and Kazuhito Mine and Tang Phu Thien Nhan and Yuga Yano and Yuichiro Tanaka and Takeshi Nishida and Takashi Morie and Hakaru Tamukoh},
title = {Hibikino-Musashi@Home 2025 Team Description Paper},
journal = {arXiv preprint arXiv:2511.20180},
year = {2025},
doi = {10.48550/arXiv.2511.20180}
}
Contact
Hibikino-Musashi@Home Kyushu Institute of Technology / The University of Kitakyushu hma@brain.kyutech.ac.jp https://www.brain.kyutech.ac.jp/~hma/