Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

UniGeo: A Unified 3D Indoor Object Detection Framework Integrating Geometry-Aware Learning and Dynamic Channel Gating

Xing Yi, Jinyang Huang, Feng-Qi Cui, Anyang Tong, Ruimin Wang, Liu Liu, Dan Guo

TL;DR

UniGeo addresses indoor 3D object detection from sparse point clouds by explicitly modeling geometric relationships and adaptively calibrating channel features. It introduces a geometry-aware learning module that maps spatial structure to feature weights via a centroid-based Euclidean distance and an exponential decay, and a dynamic channel gating mechanism that learns per-channel importance to enhance discriminative geometric cues. A hybrid feature aggregation combines global geometric calibration with local channel-enhanced details, feeding a transformer encoder to produce 3D bounding boxes and classes, trained with a UniDet3D-like loss. Experiments across six indoor datasets show state-of-the-art performance and strong generalization, with ablation studies confirming the complementary benefits of GAL and DCG and the effectiveness of Euclidean-distance-based geometry weighting.

Abstract

The growing adoption of robotics and augmented reality in real-world applications has driven considerable research interest in 3D object detection based on point clouds. While previous methods address unified training across multiple datasets, they fail to model geometric relationships in sparse point cloud scenes and ignore the feature distribution in significant areas, which ultimately restricts their performance. To deal with this issue, a unified 3D indoor detection framework, called UniGeo, is proposed. To model geometric relations in scenes, we first propose a geometry-aware learning module that establishes a learnable mapping from spatial relationships to feature weights, which enabes explicit geometric feature enhancement. Then, to further enhance point cloud feature representation, we propose a dynamic channel gating mechanism that leverages learnable channel-wise weighting. This mechanism adaptively optimizes features generated by the sparse 3D U-Net network, significantly enhancing key geometric information. Extensive experiments on six different indoor scene datasets clearly validate the superior performance of our method.

UniGeo: A Unified 3D Indoor Object Detection Framework Integrating Geometry-Aware Learning and Dynamic Channel Gating

TL;DR

UniGeo addresses indoor 3D object detection from sparse point clouds by explicitly modeling geometric relationships and adaptively calibrating channel features. It introduces a geometry-aware learning module that maps spatial structure to feature weights via a centroid-based Euclidean distance and an exponential decay, and a dynamic channel gating mechanism that learns per-channel importance to enhance discriminative geometric cues. A hybrid feature aggregation combines global geometric calibration with local channel-enhanced details, feeding a transformer encoder to produce 3D bounding boxes and classes, trained with a UniDet3D-like loss. Experiments across six indoor datasets show state-of-the-art performance and strong generalization, with ablation studies confirming the complementary benefits of GAL and DCG and the effectiveness of Euclidean-distance-based geometry weighting.

Abstract

The growing adoption of robotics and augmented reality in real-world applications has driven considerable research interest in 3D object detection based on point clouds. While previous methods address unified training across multiple datasets, they fail to model geometric relationships in sparse point cloud scenes and ignore the feature distribution in significant areas, which ultimately restricts their performance. To deal with this issue, a unified 3D indoor detection framework, called UniGeo, is proposed. To model geometric relations in scenes, we first propose a geometry-aware learning module that establishes a learnable mapping from spatial relationships to feature weights, which enabes explicit geometric feature enhancement. Then, to further enhance point cloud feature representation, we propose a dynamic channel gating mechanism that leverages learnable channel-wise weighting. This mechanism adaptively optimizes features generated by the sparse 3D U-Net network, significantly enhancing key geometric information. Extensive experiments on six different indoor scene datasets clearly validate the superior performance of our method.
Paper Structure (13 sections, 9 equations, 2 figures, 6 tables)

This paper contains 13 sections, 9 equations, 2 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Method
  3. Approach Overview
  4. Geometry-aware Learning
  5. Dynamic Channel Gating Mechanism
  6. Feature Aggregation
  7. Transformer Encoder and Loss
  8. Experimental
  9. Experimental Setup
  10. Analysis of Results
  11. Ablation Study
  12. Conclusion
  13. ACKNOWLEDGEMENTS

Figures (2)

  • Figure 1: Overview of our method. UniGeo takes a point cloud as input and pass through a geometry-aware learning module and a dynamic channel gating mechanism to generate geometry weights and channel features. A feature aggregator combines these with superpoint features into a hybrid representation, which then serves as input queries to a transformer encoder. Finally, a box MLP and a class MLP predict the 3D bounding boxes from the transformer encoder outputs.
  • Figure 2: Qualitative comparison on six indoor scenes datasets.