IPGN: Interactiveness Proposal Graph Network for Human-Object Interaction Detection.

Area of Science:

• Computer Vision
• Artificial Intelligence

Background:

• Human-Object Interaction (HOI) detection is crucial for understanding human actions with objects.
• Existing methods often treat HOI detection as a complex triplet classification problem.
• A need exists for more efficient and effective HOI detection models.

Purpose of the Study:

• To propose a novel two-stage graph model for Human-Object Interaction (HOI) detection.
• To decompose HOI detection into learning interactiveness and interaction.
• To develop the Interactiveness Proposal Graph Network (IPGN).

Main Methods:

• A two-stage graph network approach is proposed: Interactiveness Proposal Graph Network (IPGN).
• Stage one uses a fully connected graph to learn class-agnostic interactiveness, identifying interactive human-object pairs.
• Stage two employs a sparsely connected graph, guided by interactiveness knowledge, for detailed interaction learning.

Main Results:

• The proposed IPGN model achieves state-of-the-art performance on benchmark datasets.
• The two-stage approach effectively learns both interactiveness and interaction knowledge.
• Class-agnostic interactiveness provides effective proposals for interaction learning.

Conclusions:

• The novel two-stage graph model significantly advances HOI detection.
• IPGN demonstrates superior performance by effectively modeling interactiveness and interaction.
• This approach offers a more general and efficient solution for HOI detection tasks.