RealX3D: A Physically-Degraded 3D Benchmark for Multi-view Visual Restoration and Reconstruction

2025-12-29

AI Summary

Existing multi-view 3D reconstruction benchmarks lack rigorous evaluation under realistic physical degradations—such as non-uniform illumination, atmospheric scattering, occlusions, and motion blur—limiting assessment of algorithmic robustness. Method: We introduce RealX3D, the first multi-view 3D benchmark explicitly designed for realistic degradations. It leverages synchronized RGB, RAW, and LiDAR data acquisition, physics-driven degradation control, and pixel-level registration to generate world-scale ground-truth LiDAR meshes, metric depth maps, and paired low-quality/ground-truth images. Crucially, RealX3D incorporates the first multi-level severity annotations for degradation types. Results: Experiments reveal substantial performance degradation—up to 42% in Chamfer distance—of state-of-the-art optimization- and feedforward-based reconstruction methods under realistic conditions. RealX3D establishes a reproducible, quantifiable benchmark for robust multi-view vision, bridging the critical gap between realistic physical degradation modeling and systematic, empirical evaluation.

Technology Category

• Computer Vision: 3D Computer Vision
• Machine Learning: Multi-instance/Multi-view Learning
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets
• Security and Privacy: Large-scale security measurements

Abstract

We introduce RealX3D, a real-capture benchmark for multi-view visual restoration and 3D reconstruction under diverse physical degradations. RealX3D groups corruptions into four families, including illumination, scattering, occlusion, and blurring, and captures each at multiple severity levels using a unified acquisition protocol that yields pixel-aligned LQ/GT views. Each scene includes high-resolution capture, RAW images, and dense laser scans, from which we derive world-scale meshes and metric depth. Benchmarking a broad range of optimization-based and feed-forward methods shows substantial degradation in reconstruction quality under physical corruptions, underscoring the fragility of current multi-view pipelines in real-world challenging environments.

Problem

• Develops a benchmark for multi-view restoration and reconstruction under physical degradations
• Groups corruptions into four families with multiple severity levels
• Evaluates current methods' fragility in real-world challenging environments

Innovation

• RealX3D benchmark groups corruptions into four families
• Unified protocol yields pixel-aligned LQ/GT views
• Includes high-resolution capture, RAW images, laser scans