Unsupervised Specialization of Visual Subclasses Using K-Means in YOLO-Based Detection Pipelines

• Pedro Henrique Campos Moreira UFV
• Bianca Panacho Ferreira UFV
• Marcus Vinicius Diniz dos Reis UFV

DOI: https://doi.org/10.5753/wsis.2025.15733

Abstract

This work presents a pipeline for enriching object detection datasets through automatic visual subclass labeling. Using generic annotations, the methodology uses a YOLO detector to extract object instances and then applies image processing to extract color vectors. The unsupervised K-Means algorithm is used to cluster these vectors, autonomously defining new subclasses. The YOLOv8n model, retrained with the refined dataset, achieved 94.68% accuracy in distinguishing sports teams, validating the approach as an effective solution for overcoming the need for manual annotation.

References

Grishin, K., Chupin, S., Vasylenko, A., Barkhatova, T., and Burenin, R. (2023). Yolocl: Galaxy cluster detection in the sdss with deep machine learning. Astronomy & Astrophysics, 677:A101.

Koshkina, A., Nauata, N., Tighe, J., and Felsen, P. (2021). Unsupervised classification of players in team sports. arXiv preprint arXiv:2104.10068.

Kowsari, K. and Alassaf, M. H. (2016). Weighted unsupervised learning for 3d object detection. arXiv preprint arXiv:1602.05920.

LeCun, Y., Bottou, L., Bengio, Y., and Haffner, P. (2002). Gradient-based learning applied to document recognition. Proceedings of the IEEE, 86(11):2278–2324.

MacQueen, J. (1967). Multivariate observations. In Proceedings ofthe 5th Berkeley Symposium on Mathematical Statisticsand Probability, volume 1, pages 281–297.

Qiu, T., Yang, L., Zhang, L., Tang, J., and Xu, X. (2024). Clda-yolo: Visual contrastive learning based domain adaptive yolo detector. arXiv preprint arXiv:2412.11812.

Redmon, J. and Farhadi, A. (2017). Yolo9000: better, faster, stronger. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 7263–7271.