PRISM: Feed-Forward Single-Image 3D Reconstruction via Geometric Warp-Residual Modeling
This work addresses the efficiency bottleneck of diffusion-based 3D reconstruction for practitioners needing fast inference in VR, robotics, and content creation.
PRISM introduces a feed-forward framework for single-image 3D reconstruction that decomposes multi-view latent prediction into geometric warping and residual correction, eliminating iterative diffusion sampling. It achieves competitive quality with diffusion-based methods while reducing inference time to 36 seconds per scene.
Reconstructing 3D scenes from a single image is a fundamental challenge in computer vision, with broad applications in virtual reality, robotics, and content creation. Recent methods achieve outstanding performance by leveraging camera-controlled video diffusion models, but rely on iterative diffusion sampling, which greatly limits their practical deployment. We observe that geometric forward warping alone can cover the majority of a target view directly from the input image, with only a compact residual left for the encoder to correct. Motivated by this observation, we propose PRISM, a feed-forward framework that decomposes multi-view latent prediction into a parameter-free geometric prior and a learned residual correction, with no diffusion sampling required at inference. To enable generalization from purely synthetic training data, we devise a two-stage training strategy combining latents supervised distillation for geometric generalization and perceptual fine-tuning for appearance quality optimization. Extensive experiments on three benchmarks demonstrate that PRISM achieves competitive reconstruction quality compared with diffusion-based methods, while reducing inference time dramatically to only 36 seconds per scene.