RefRef: A Synthetic Dataset and Benchmark for Reconstructing Refractive and Reflective Objects

2025-05-09

AI Summary

Existing 3D reconstruction and novel view synthesis methods predominantly assume Lambertian reflectance and rectilinear light propagation, rendering them inadequate for modeling nonlinear optical phenomena such as refraction and reflection; moreover, no standardized benchmark exists for evaluating performance under complex optical effects. Method: We introduce RefRef—the first synthetic dataset (150 scenes) specifically designed for complex optical phenomena—and establish a standardized evaluation protocol. Technically, we propose an oracle neural renderer grounded in physically accurate light transport simulation, alongside a differentiable surrogate that requires neither prior geometry nor refractive index knowledge, integrating ray tracing, PBRT-based rendering, and NeRF extensions. Results: Experiments reveal that current state-of-the-art methods underperform the oracle by over 8.2 dB in PSNR on RefRef, confirming the task’s difficulty and establishing the first systematic evaluation baseline and methodological framework for non-Lambertian scene reconstruction.

Technology Category

• Computer Vision: Computational Photography, Image & Video Synthesis
• Search and Optimization: Sampling/Simulation-based Search
• 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
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

Modern 3D reconstruction and novel view synthesis approaches have demonstrated strong performance on scenes with opaque Lambertian objects. However, most assume straight light paths and therefore cannot properly handle refractive and reflective materials. Moreover, datasets specialized for these effects are limited, stymieing efforts to evaluate performance and develop suitable techniques. In this work, we introduce a synthetic RefRef dataset and benchmark for reconstructing scenes with refractive and reflective objects from posed images. Our dataset has 50 such objects of varying complexity, from single-material convex shapes to multi-material non-convex shapes, each placed in three different background types, resulting in 150 scenes. We also propose an oracle method that, given the object geometry and refractive indices, calculates accurate light paths for neural rendering, and an approach based on this that avoids these assumptions. We benchmark these against several state-of-the-art methods and show that all methods lag significantly behind the oracle, highlighting the challenges of the task and dataset.

Problem

• Handling refractive and reflective materials in 3D reconstruction
• Lack of specialized datasets for evaluating such techniques
• Developing accurate light path methods for neural rendering

Innovation

• Synthetic dataset for refractive and reflective objects
• Oracle method calculates accurate light paths
• Benchmark compares state-of-the-art reconstruction methods