SAF3R: Dynamic Sparse Attention for Feed-Forward 3D Reconstruction Transformers

2026-07-03

AI Summary

This work addresses the computational bottleneck in feedforward 3D reconstruction Transformers caused by the quadratic complexity of global attention when processing long sequences. To overcome this limitation, the authors propose a training-free dynamic sparse attention framework that leverages the observed high heterogeneity, dynamism, and extreme sparsity of attention patterns in F3R Transformers. The approach integrates offline attention head profiling, input-aware online sparse mask generation, and efficient sparse computation to enable input-dependent adaptive sparsification. Experimental results demonstrate that the method significantly accelerates end-to-end inference while preserving accuracy in camera pose estimation and 3D reconstruction, outperforming existing sparse attention techniques.

Abstract

Feed-forward 3D reconstruction (F3R) transformers have recently achieved remarkable success. However, scaling them to long image sequences remains challenging, as the quadratic complexity of cross-view global attention quickly becomes the dominant computational bottleneck. While recent efforts attempt to improve efficiency through compressed or sparse attention, they fail to fully exploit the inherent sparsity and dynamic behavior of global attention. In this work, we present a comprehensive analysis of global attention across multiple F3R transformers and reveal that attention patterns are highly heterogeneous, dynamic, and extremely sparse across layers and attention heads. Motivated by these findings, we propose SAF3R, a training-free dynamic sparse attention framework tailored to F3R transformers. SAF3R integrates tailored sparse attention mechanisms with offline head profiling and an efficient online adaptation strategy to match input-dependent attention behaviors. Extensive experiments demonstrate that SAF3R achieves high sparsity ratios while preserving camera pose estimation and 3D reconstruction quality, translating into substantial end-to-end speedup on F3R transformers compared to existing methods. Code is available at https://github.com/jndeng/SAF3R

Problem

• Feed-forward 3D reconstruction
• global attention
• computational bottleneck
• attention sparsity
• dynamic attention

Innovation

• dynamic sparse attention
• feed-forward 3D reconstruction
• training-free
• attention sparsity
• transformer acceleration