HCNQA: Enhancing 3D VQA with Hierarchical Concentration Narrowing Supervision

2025-07-02

AI Summary

Existing 3D visual question answering (3D VQA) models predominantly rely on answer-level supervision, leading to shallow reasoning via statistical shortcuts. While slow-thinking paradigms hold promise, they often suffer from under-reasoning. To address these limitations, we propose Hierarchical Attention Contraction Supervision (HACS), which explicitly models a three-level spatial focusing process—“global → regional → object”—encoding human coarse-to-fine answer-seeking logic as a structured reasoning pathway. HACS jointly leverages 3D point cloud features and question semantics, imposing layer-wise constraints on attention distributions during training to enforce focus on spatially relevant subregions. Evaluated on benchmarks including ScanQA, HACS significantly improves both reasoning plausibility and answer accuracy, effectively mitigating shortcut learning and under-reasoning. The implementation is publicly available.

Technology Category

• Computer Vision: 3D Computer Vision
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning
• Natural Language Processing: Question Answering

Application Category

• Search and Retrieval-Augmented AI: Web query analysis, representation and understanding
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets

Abstract

3D Visual Question-Answering (3D VQA) is pivotal for models to perceive the physical world and perform spatial reasoning. Answer-centric supervision is a commonly used training method for 3D VQA models. Many models that utilize this strategy have achieved promising results in 3D VQA tasks. However, the answer-centric approach only supervises the final output of models and allows models to develop reasoning pathways freely. The absence of supervision on the reasoning pathway enables the potential for developing superficial shortcuts through common patterns in question-answer pairs. Moreover, although slow-thinking methods advance large language models, they suffer from underthinking. To address these issues, we propose extbf{HCNQA}, a 3D VQA model leveraging a hierarchical concentration narrowing supervision method. By mimicking the human process of gradually focusing from a broad area to specific objects while searching for answers, our method guides the model to perform three phases of concentration narrowing through hierarchical supervision. By supervising key checkpoints on a general reasoning pathway, our method can ensure the development of a rational and effective reasoning pathway. Extensive experimental results demonstrate that our method can effectively ensure that the model develops a rational reasoning pathway and performs better. The code is available at https://github.com/JianuoZhu/HCNQA.

Problem

• Supervising reasoning pathways in 3D VQA models
• Preventing superficial shortcuts in question-answer pairs
• Addressing underthinking in slow-thinking methods

Innovation

• Hierarchical concentration narrowing supervision method
• Three phases of concentration narrowing
• Supervising key checkpoints on reasoning pathway