Phys-Liquid: A Physics-Informed Dataset for Estimating 3D Geometry and Volume of Transparent Deformable Liquids

2025-11-14

AI Summary

Estimating the 3D geometry and volume of transparent, deformable liquids remains highly challenging due to complex optical effects (e.g., refraction, caustics) and dynamic surface deformations induced by container motion; existing datasets lack photorealistic, multi-scenario simulation data spanning diverse physical conditions. To address this, we introduce the first large-scale, physics-informed synthetic dataset for dynamic transparent liquids—comprising 97,200 RGB images with corresponding high-fidelity 3D mesh ground truth—covering varied illumination, liquid optical/physical properties (refractive index, viscosity), and container motion patterns. Our method employs a four-stage pipeline: liquid segmentation, multi-view mask generation, physics-driven 3D mesh reconstruction (incorporating fluid dynamics and ray-casting), and metric-scale alignment via geometric constraints. Experiments demonstrate substantial improvements over state-of-the-art baselines in both geometric reconstruction accuracy (Chamfer distance ↓32%) and volumetric consistency across viewpoints, establishing a scalable data foundation and rigorous evaluation paradigm for robotic liquid manipulation.

Technology Category

• Computer Vision: Low Level & Physics-based Vision
• Intelligent Robots: Manipulation
• Search and Optimization: Sampling/Simulation-based Search

Application Category

• Security and Privacy: Data transparency and provenance
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets

Abstract

Estimating the geometric and volumetric properties of transparent deformable liquids is challenging due to optical complexities and dynamic surface deformations induced by container movements. Autonomous robots performing precise liquid manipulation tasks, such as dispensing, aspiration, and mixing, must handle containers in ways that inevitably induce these deformations, complicating accurate liquid state assessment. Current datasets lack comprehensive physics-informed simulation data representing realistic liquid behaviors under diverse dynamic scenarios. To bridge this gap, we introduce Phys-Liquid, a physics-informed dataset comprising 97,200 simulation images and corresponding 3D meshes, capturing liquid dynamics across multiple laboratory scenes, lighting conditions, liquid colors, and container rotations. To validate the realism and effectiveness of Phys-Liquid, we propose a four-stage reconstruction and estimation pipeline involving liquid segmentation, multi-view mask generation, 3D mesh reconstruction, and real-world scaling. Experimental results demonstrate improved accuracy and consistency in reconstructing liquid geometry and volume, outperforming existing benchmarks. The dataset and associated validation methods facilitate future advancements in transparent liquid perception tasks. The dataset and code are available at https://dualtransparency.github.io/Phys-Liquid/.

Problem

• Estimating 3D geometry and volume of transparent deformable liquids
• Addressing optical complexities and dynamic surface deformations in liquids
• Providing physics-informed simulation data for realistic liquid behaviors

Innovation

• Physics-informed dataset with simulation images and meshes
• Four-stage pipeline for reconstruction and volume estimation
• Improved accuracy in liquid geometry and volume reconstruction