analiza aiqbench
This paper addresses the computational bottleneck of Vision Transformers by introducing ReConV, a reconstruction-guided framework that compresses visual token sequences without sacrificing critical information. By retaining only tokens with high reconstruction residuals while summarizing shared data in a compact bank, the method achieves significant efficiency gains, such as reducing GMACs by 48% on ImageNet. This approach offers a practical pathway for deploying large-scale vision models in resource-constrained environments through efficient multi-turn inference.
Bottlenecked semantic compression by representation reconstruction
arXiv:2610.17412v1 Announce Type: cross Abstract: Vision Transformers incur high computational costs when processing long visual token sequences. Existing reduction methods typically rank tokens independently and discard those deemed unimportant or redundant. We propose a reconstruction-guided framework for visual token condensation (ReConV), which summarizes the complete visual sequence with a compact Summary Bank (SMB) and retains tokens whose information is not adequately captured by the SMB, as indicated by their reconstruction residuals. At each compression layer, a lightweight position-conditioned decoder reconstructs the current tokens, leaving shared information compressed in the SMB while explicitly preserving poorly reconstructed details. Experiments on image classification and multimodal understanding demonstrate favorable performance-efficiency trade-offs: ReConV surpasses the uncompressed baseline with 48\% fewer GMACs on ImageNet and enables reusable image-level representations for efficient multi-turn inference. Code is available at TokenDrop.
Abstract:Vision Transformers incur high computational costs when processing long visual token sequences. Existing reduction methods typically rank tokens independently and discard those deemed unimportant or redundant. We propose a reconstruction-guided framework for visual token condensation (ReConV), which summarizes the complete visual sequence with a compact Summary Bank (SMB) and retains tokens whose information is not adequately captured by the SMB, as indicated by their reconstruction residuals. At each compression layer, a lightweight position-conditioned decoder reconstructs the current tokens, leaving shared information compressed in the SMB while explicitly preserving poorly reconstructed details. Experiments on image classification and multimodal understanding demonstrate favorable performance-efficiency trade-offs: ReConV surpasses the uncompressed baseline with 48\% fewer GMACs on ImageNet and enables reusable image-level representations for efficient multi-turn inference. Code is available at TokenDrop.
Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
arXiv:2610.17412 [cs.CV]
(or arXiv:2610.17412v1 [cs.CV] for this version)
https://doi.org/10.48550/arXiv.2610.17412 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 227 cuvinte
Subiecte conexe
Lossy Compressive Text Autoencoders
arXiv · 9 oct. 2026
Rethinking Causal Action Tokenization with Conditional Annealing in Flow Matching
arXiv · 30 sept. 2026
GS-Codec: A Gaussian-Splatting Bottleneck for Neural Audio Coding
arXiv · 7 oct. 2026
Recover Semantics First, Generate Better: Improved Latent Modeling for 3D MRI Reconstruction and Cross-Contrast Synthesis
arXiv · 17 iun. 2026