2026
Hu, Jianing Deng; Yuanzhe LI; Jialu Wang; Song Wang; Tianlong Chen; Huanrui Yang; Jingtong
SAF3R: Dynamic Sparse Attention for Feed-Forward 3D Reconstruction Transformers Conference
European Conference on Computer Vision -- ECCV 2026, 2026.
Abstract | Tags: 3D Reconstruction, ECCV | Links:
@conference{nokey,
title = {SAF3R: Dynamic Sparse Attention for Feed-Forward 3D Reconstruction Transformers},
author = {Jianing Deng; Yuanzhe LI; Jialu Wang; Song Wang; Tianlong Chen; Huanrui Yang; Jingtong Hu},
url = {https://arxiv.org/abs/2607.03612},
year = {2026},
date = {2026-09-07},
publisher = {European Conference on Computer Vision -- ECCV 2026},
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. },
keywords = {3D Reconstruction, ECCV},
pubstate = {published},
tppubtype = {conference}
}
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.