B-RIGHT: Benchmark Re-evaluation for Integrity in Generalized Human-Object Interaction Testing


Yoojin Jang (Ulsan National Institute of Science and Technology), Junsu Kim (Ulsan National Institute of Science and Technology), Ha Yeon Kim (Ulsan National Institute of Science and Technology), Eun-Ki Lee (Hanyang University), Eun-Sol Kim (Hanyang University), Seungryul Baek (Ulsan National Institute of Science and Technology), Jaejun Yoo (Ulsan National Institute of Science and Technology)
The 35th British Machine Vision Conference

Abstract

Human-object interaction (HOI) detection aims to understand complex relationships between humans and objects, but existing benchmarks like HICO-DET suffer from severe class imbalance and inconsistent train-test splits, compromising reliable evaluations. To address these issues, we introduce **B-RIGHT** (**B**enchmark **R**e-evaluation for **I**ntegrity in **G**eneralized **H**uman-Object Interaction **T**esting), a systematically balanced dataset constructed via a novel balancing algorithm and an automated image generation-and-filtering pipeline. By ensuring uniform representation across HOI classes, B-RIGHT significantly reduces performance variance. Our re-evaluation of state-of-the-art methods reveals substantial shifts in model rankings compared to HICO-DET, highlighting previously hidden biases. These results underscore the critical importance of balanced benchmarks for fair and insightful model comparisons. The dataset is publicly available at [B-RIGHT](https://github.com/hellog2n/B-RIGHT).

Citation

@inproceedings{Jang_2025_BMVC,
author    = {Yoojin Jang and Junsu Kim and Ha Yeon Kim and Eun-Ki Lee and Eun-Sol Kim and Seungryul Baek and Jaejun Yoo},
title     = {B-RIGHT: Benchmark Re-evaluation for Integrity in Generalized Human-Object Interaction Testing},
booktitle = {36th British Machine Vision Conference 2025, {BMVC} 2025, Sheffield, UK, November 24-27, 2025},
publisher = {BMVA},
year      = {2025},
url       = {https://bmva-archive.org.uk/bmvc/2025/assets/papers/Paper_272/paper.pdf}
}


Copyright © 2025 The British Machine Vision Association and Society for Pattern Recognition
The British Machine Vision Conference is organised by The British Machine Vision Association and Society for Pattern Recognition. The Association is a Company limited by guarantee, No.2543446, and a non-profit-making body, registered in England and Wales as Charity No.1002307 (Registered Office: Dept. of Computer Science, Durham University, South Road, Durham, DH1 3LE, UK).

Imprint | Data Protection