Saved in:
Bibliographic Details
Main Authors: Li, Huan, Luo, Longjun, Shi, Yuling, Gu, Xiaodong
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2512.21691
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911339157389312
author Li, Huan
Luo, Longjun
Shi, Yuling
Gu, Xiaodong
author_facet Li, Huan
Luo, Longjun
Shi, Yuling
Gu, Xiaodong
contents Visual Geometry Grounded Transformer (VGGT) delivers state-of-the-art feed-forward 3D reconstruction, yet its global self-attention layer suffers from a drastic collapse phenomenon when the input sequence exceeds a few hundred frames: attention matrices rapidly become near rank-one, token geometry degenerates to an almost one-dimensional subspace, and reconstruction error accumulates super-linearly.In this report,we establish a rigorous mathematical explanation of the collapse by viewing the global-attention iteration as a degenerate diffusion process.We prove that,in VGGT, the token-feature flow converges toward a Dirac-type measure at a $O(1/L)$ rate, where $L$ is the layer index, yielding a closed-form mean-field partial differential equation that precisely predicts the empirically observed rank profile.The theory quantitatively matches the attention-heat-map evolution and a series of experiments outcomes reported in relevant works and explains why its token-merging remedy -- which periodically removes redundant tokens -- slows the effective diffusion coefficient and thereby delays collapse without additional training.We believe the analysis provides a principled lens for interpreting future scalable 3D-vision transformers,and we highlight its potential for multi-modal generalization.
format Preprint
id arxiv_https___arxiv_org_abs_2512_21691
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Analyzing the Mechanism of Attention Collapse in VGGT from a Dynamics Perspective
Li, Huan
Luo, Longjun
Shi, Yuling
Gu, Xiaodong
Computer Vision and Pattern Recognition
Visual Geometry Grounded Transformer (VGGT) delivers state-of-the-art feed-forward 3D reconstruction, yet its global self-attention layer suffers from a drastic collapse phenomenon when the input sequence exceeds a few hundred frames: attention matrices rapidly become near rank-one, token geometry degenerates to an almost one-dimensional subspace, and reconstruction error accumulates super-linearly.In this report,we establish a rigorous mathematical explanation of the collapse by viewing the global-attention iteration as a degenerate diffusion process.We prove that,in VGGT, the token-feature flow converges toward a Dirac-type measure at a $O(1/L)$ rate, where $L$ is the layer index, yielding a closed-form mean-field partial differential equation that precisely predicts the empirically observed rank profile.The theory quantitatively matches the attention-heat-map evolution and a series of experiments outcomes reported in relevant works and explains why its token-merging remedy -- which periodically removes redundant tokens -- slows the effective diffusion coefficient and thereby delays collapse without additional training.We believe the analysis provides a principled lens for interpreting future scalable 3D-vision transformers,and we highlight its potential for multi-modal generalization.
title Analyzing the Mechanism of Attention Collapse in VGGT from a Dynamics Perspective
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2512.21691