Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Syn-to-Real Unsupervised Domain Adaptation for Indoor 3D Object Detection

Yunsong Wang, Na Zhao, Gim Hee Lee

TL;DR

This work addresses the challenge of syn-to-real unsupervised domain adaptation for indoor 3D object detection by introducing Object-wise Hierarchical Domain Alignment (OHDA). OHDA combines object-aware augmentation with a two-branch adaptation framework that jointly performs class-level pseudo-label guided alignment and holistic proposal-level domain discrimination, complemented by progressive pseudo-label refinement and perturbation-based reweighting. The approach achieves substantial mAP improvements over Source-Only baselines on 3D-FRONT → ScanNetV2 and 3D-FRONT → SUN RGB-D benchmarks (approximately 9–10 percentage points), outperforming methods adapted from outdoor contexts. These results demonstrate the practicality of syn-to-real indoor UDA and establish a strong benchmark for future indoor domain adaptation research, with code to be released upon acceptance.

Abstract

The use of synthetic data in indoor 3D object detection offers the potential of greatly reducing the manual labor involved in 3D annotations and training effective zero-shot detectors. However, the complicated domain shifts across syn-to-real indoor datasets remains underexplored. In this paper, we propose a novel Object-wise Hierarchical Domain Alignment (OHDA) framework for syn-to-real unsupervised domain adaptation in indoor 3D object detection. Our approach includes an object-aware augmentation strategy to effectively diversify the source domain data, and we introduce a two-branch adaptation framework consisting of an adversarial training branch and a pseudo labeling branch, in order to simultaneously reach holistic-level and class-level domain alignment. The pseudo labeling is further refined through two proposed schemes specifically designed for indoor UDA. Our adaptation results from synthetic dataset 3D-FRONT to real-world datasets ScanNetV2 and SUN RGB-D demonstrate remarkable mAP25 improvements of 9.7% and 9.1% over Source-Only baselines, respectively, and consistently outperform the methods adapted from 2D and 3D outdoor scenarios. The code will be publicly available upon paper acceptance.

Syn-to-Real Unsupervised Domain Adaptation for Indoor 3D Object Detection

TL;DR

This work addresses the challenge of syn-to-real unsupervised domain adaptation for indoor 3D object detection by introducing Object-wise Hierarchical Domain Alignment (OHDA). OHDA combines object-aware augmentation with a two-branch adaptation framework that jointly performs class-level pseudo-label guided alignment and holistic proposal-level domain discrimination, complemented by progressive pseudo-label refinement and perturbation-based reweighting. The approach achieves substantial mAP improvements over Source-Only baselines on 3D-FRONT → ScanNetV2 and 3D-FRONT → SUN RGB-D benchmarks (approximately 9–10 percentage points), outperforming methods adapted from outdoor contexts. These results demonstrate the practicality of syn-to-real indoor UDA and establish a strong benchmark for future indoor domain adaptation research, with code to be released upon acceptance.

Abstract

The use of synthetic data in indoor 3D object detection offers the potential of greatly reducing the manual labor involved in 3D annotations and training effective zero-shot detectors. However, the complicated domain shifts across syn-to-real indoor datasets remains underexplored. In this paper, we propose a novel Object-wise Hierarchical Domain Alignment (OHDA) framework for syn-to-real unsupervised domain adaptation in indoor 3D object detection. Our approach includes an object-aware augmentation strategy to effectively diversify the source domain data, and we introduce a two-branch adaptation framework consisting of an adversarial training branch and a pseudo labeling branch, in order to simultaneously reach holistic-level and class-level domain alignment. The pseudo labeling is further refined through two proposed schemes specifically designed for indoor UDA. Our adaptation results from synthetic dataset 3D-FRONT to real-world datasets ScanNetV2 and SUN RGB-D demonstrate remarkable mAP25 improvements of 9.7% and 9.1% over Source-Only baselines, respectively, and consistently outperform the methods adapted from 2D and 3D outdoor scenarios. The code will be publicly available upon paper acceptance.
Paper Structure (19 sections, 7 equations, 4 figures, 3 tables)

This paper contains 19 sections, 7 equations, 4 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Our Approach
  4. Overview
  5. Cross-Domain Data Alignment
  6. Object-wise Hierarchical Domain Alignment
  7. Class-Level Alignment
  8. Holistic-Level Alignment
  9. Pseudo Label Refinement
  10. Progressive Class-Aware Thresholding
  11. Model Perturbation based Reweighting
  12. Training Objective
  13. Experiments
  14. Implementations Details
  15. Baselines
  16. ...and 4 more sections

Figures (4)

  • Figure 1: Pipelines of Source-Only and OHDA.
  • Figure 2: Proposed Framework. Our framework consists of three parts: cross-domain data alignment, Object-wise Hierarchical Domain Alignment (consisting of Class-Level Alignment (CLA) and Holistic-Level Alignment (HLA)), and Pseudo Label Refinement (PLR).
  • Figure 3: Qualitative results on ScanNetV2 (first row) and SUN RGB-D (second row).
  • Figure 4: Qualitative results for ablation study. Better viewed in color with zoom in.