Holistic Interaction Transformer Network for Action Detection

23 Oct 2022  ·  Gueter Josmy Faure, Min-Hung Chen, Shang-Hong Lai ·

Actions are about how we interact with the environment, including other people, objects, and ourselves. In this paper, we propose a novel multi-modal Holistic Interaction Transformer Network (HIT) that leverages the largely ignored, but critical hand and pose information essential to most human actions. The proposed "HIT" network is a comprehensive bi-modal framework that comprises an RGB stream and a pose stream. Each of them separately models person, object, and hand interactions. Within each sub-network, an Intra-Modality Aggregation module (IMA) is introduced that selectively merges individual interaction units. The resulting features from each modality are then glued using an Attentive Fusion Mechanism (AFM). Finally, we extract cues from the temporal context to better classify the occurring actions using cached memory. Our method significantly outperforms previous approaches on the J-HMDB, UCF101-24, and MultiSports datasets. We also achieve competitive results on AVA. The code will be available at https://github.com/joslefaure/HIT.

PDF Abstract

Results from the Paper

Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Action Recognition AVA v2.2 HIT mAP 32.6 # 22
Action Detection J-HMDB HIT Video-mAP 0.2 89.7 # 1
Video-mAP 0.5 88.1 # 1
Frame-mAP 0.5 83.8 # 1
Action Detection MultiSports HIT Video-mAP 0.2 27.8 # 1
Video-mAP 0.5 8.8 # 1
Frame-mAP 0.5 33.3 # 1
Action Detection UCF101-24 HIT Video-mAP 0.2 88.8 # 1
Video-mAP 0.5 74.3 # 1
Frame-mAP 0.5 84.8 # 3