GHOI: A Green Human-Object-Interaction Detector

Tsung-Shan Yang, Yun-Cheng Wang, Chengwei Wei, C.-C. Jay Kuo

Abstract: Human-Object Interaction (HOI) detection is a fundamental task in image understanding. All recent high- performance HOI methods are based on deep learning (DL) models, which are computationally expensive with an opaque inference process. A green HOI (GHOI) detector is proposed in this work to strike a good balance between detection per-formance, inference complexity (i.e., low carbon footprints), and mathematical transparency. GHOI is a two-stage method. In the first stage, it conducts object detection and extracts various features from the input images as intermediate outputs. In the second stage, it uses the first-stage outputs to predict the interaction type using the XGBoost classifier. One novel contribution is the application of error correction codes (ECCs) to encode rare interaction cases. This reduces the model size and the complexity of the XGBoost classifier in the second stage. Experimental results demonstrate the advantages of ECC-coded interaction labels and the nice balance of detection performance and complexity of the proposed GHOI method.