ERNet: An Efficient and Reliable Human-Object Interaction Detection Network.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Robotics

Background:

• Human-Object Interaction (HOI) detection is crucial for autonomous systems like self-driving cars and robots.
• Existing HOI detectors face challenges with efficiency and prediction reliability, limiting real-world applications.

Purpose of the Study:

• To address the limitations of current HOI detectors by proposing an efficient and reliable end-to-end trainable network.
• To enhance the accuracy and robustness of HOI detection in complex scenarios.

Main Methods:

• Introduced ERNet, a convolutional-transformer network featuring efficient multi-scale deformable attention for HOI feature extraction.
• Developed a novel detection attention module for generating semantically rich instance and interaction tokens, enhancing feature refinement.
• Integrated a predictive uncertainty estimation framework for reliable HOI classification.

Main Results:

• ERNet achieves state-of-the-art performance on HICO-Det, V-COCO, and HOI-A datasets.
• Demonstrated significant improvements in both detection accuracy and training efficiency compared to existing methods.
• The uncertainty estimation framework enables reliable HOI predictions in challenging conditions.

Conclusions:

• ERNet offers a robust and efficient solution for HOI detection, overcoming previous limitations.
• The proposed model shows strong potential for deployment in real-world autonomous systems.
• Publicly available code facilitates further research and development in HOI detection.