Detecting Human-Object Interactions with Action Co-occurrence Priors
Abstract
A common problem in human-object interaction (HOI) detection task is that numerous HOI classes have only a small number of labeled examples, resulting in training sets with a long-tailed distribution. The lack of positive labels can lead to low classification accuracy for these classes. Towards addressing this issue, we observe that there exist natural correlations and anti-correlations among human-object interactions. In this paper, we model the correlations as action co-occurrence matrices and present techniques to learn these priors and leverage them for more effective training, especially on rare classes. The utility of our approach is demonstrated experimentally, where the performance of our approach exceeds the state-of-the-art methods on both of the two leading HOI detection benchmark datasets, HICO-Det and V-COCO.
Presentation Materials
Action Co-occurrence Priors
Hierarchical Architecture
Qualitative Results
HOI probability before and after applying the projection.
Awards
Silver Prize, 26th Samsung Humantech Paper Awards (Top 1.6%)
"Detecting Human-Object Interactions with Action Co-occurrence Prior"
Samsung Electronics Co., Ltd.
Bibtex
@inproceedings{kim2020detecting,
title={Detecting human-object interactions with action co-occurrence priors},
author={Kim, Dong-Jin and Sun, Xiao and Choi, Jinsoo and Lin, Stephen and Kweon, In So},
booktitle={European Conference on Computer Vision},
pages={718--736},
year={2020},
organization={Springer}
}