SLAT-Phys predicts spatially varying material property fields directly from single RGB images with a 120x speedup.
March 26, 2026
Original Paper
SLAT-Phys: Fast Material Property Field Prediction from Structured 3D Latents
arXiv · 2603.23973
The Takeaway
It bypasses the expensive 3D reconstruction and voxelization steps usually required for physics-based simulation. This allows robots or digital twin systems to estimate Young's modulus and density in real-time (9.9s) from simple visual inputs.
From the abstract
Estimating the material property field of 3D assets is critical for physics-based simulation, robotics, and digital twin generation. Existing vision-based approaches are either too expensive and slow or rely on 3D information. We present SLAT-Phys, an end-to-end method that predicts spatially varying material property fields of 3D assets directly from a single RGB image without explicit 3D reconstruction. Our approach leverages spatially organised latent features from a pretrained 3D asset gener