analiza aiqbench
The Large Sparse Reconstruction Model (LSRM) advances feed-forward 3D reconstruction by leveraging scaled transformer context windows to recover fine-grained textures that previously required dense-view optimization. By integrating sparse attention mechanisms with a coarse-to-fine pipeline and explicit geometric routing, the model handles significantly more object tokens while maintaining high fidelity in novel view synthesis and inverse rendering tasks. This work demonstrates how architectural scaling directly improves reconstruction quality without relying on computationally expensive iterative training methods.
LSRM: High-Fidelity Object-Centric Reconstruction via Scaled Context Windows
arXiv:2604.05182v3 Announce Type: replace-cross Abstract: We introduce the Large Sparse Reconstruction Model to study how scaling transformer context windows affects feed-forward 3D reconstruction. Although recent object-centric feed-forward methods produce robust, high-quality reconstructions, they still lag behind dense-view optimization in recovering fine-grained texture and appearance. We show that expanding the context window -- by substantially increasing the number of active object and image tokens -- narrows this gap and enables high-fidelity 3D object reconstruction and inverse rendering. To scale effectively, we adapt native sparse attention for 3D reconstruction with three key contributions: (1) an efficient coarse-to-fine pipeline that focuses computation on informative regions by predicting sparse high-resolution residuals; (2) a 3D-aware spatial routing mechanism that establishes accurate 2D-3D correspondences using explicit geometric distances rather than standard attention scores; and (3) a custom block-aware sequence-parallel strategy with an All-gather-KV protocol to balance dynamic, sparse workloads across GPUs. As a result, LSRM handles 20x more object tokens and >2x more image tokens than prior state-of-the-art (SOTA) methods. Extensive evaluations on standard novel-view synthesis benchmarks show substantial gains over the current SOTA, yielding >2.4dB higher PSNR and >40% lower LPIPS. Furthermore, when extending LSRM to inverse rendering, qualitative and quantitative evaluations on widely used benchmarks demonstrate consistent improvements in texture and geometry details, achieving an LPIPS that matches or exceeds that of SOTA dense-view optimization methods. Code and model weights are available on our project page.
Abstract:We introduce the Large Sparse Reconstruction Model to study how scaling transformer context windows affects feed-forward 3D reconstruction. Although recent object-centric feed-forward methods produce robust, high-quality reconstructions, they still lag behind dense-view optimization in recovering fine-grained texture and appearance. We show that expanding the context window -- by substantially increasing the number of active object and image tokens -- narrows this gap and enables high-fidelity 3D object reconstruction and inverse rendering. To scale effectively, we adapt native sparse attention for 3D reconstruction with three key contributions: (1) an efficient coarse-to-fine pipeline that focuses computation on informative regions by predicting sparse high-resolution residuals; (2) a 3D-aware spatial routing mechanism that establishes accurate 2D-3D correspondences using explicit geometric distances rather than standard attention scores; and (3) a custom block-aware sequence-parallel strategy with an All-gather-KV protocol to balance dynamic, sparse workloads across GPUs. As a result, LSRM handles 20x more object tokens and >2x more image tokens than prior state-of-the-art (SOTA) methods. Extensive evaluations on standard novel-view synthesis benchmarks show substantial gains over the current SOTA, yielding >2.4dB higher PSNR and >40% lower LPIPS. Furthermore, when extending LSRM to inverse rendering, qualitative and quantitative evaluations on widely used benchmarks demonstrate consistent improvements in texture and geometry details, achieving an LPIPS that matches or exceeds that of SOTA dense-view optimization methods. Code and model weights are available on our project page.
Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
arXiv:2604.05182 [cs.CV]
(or arXiv:2604.05182v3 [cs.CV] for this version)
Text integral sub licența deschisă a publicației · 244 cuvinte
Subiecte conexe
BRICS Draws a New Global Line: Full Text of New Delhi Declaration
slguardian.org · 12 sept. 2026
News Archive 2026
Quantum Computing Report · 25 iul. 2026
GASE: Gaussian Splatting-Based Automated System for Reconstructing Embodied-Simulation Environments
arXiv · 17 iun. 2026
Causally Debiased Latent Action Model for Embodied Action-Conditioned World Models
arXiv · 29 sept. 2026