The paper introduces Denoising Attention (DnA), a modification to the standard softmax activation used in multihead attention for visual perception tasks. DnA employs a positive query to highlight correct class features and a negative query to suppress irrelevant features, projecting these interactions into two distinct subspaces to enhance discriminability. Experiments with a ViT-B backbone show an absolute 0.8% improvement on ImageNet-1K and additional gains on video understanding and video LLM tasks.
arXiv:2604. 00757v2 Announce Type: replace-cross Abstract: Large Vision Language Models show impressive performance across image and video understanding tasks, yet their computational cost grows rapidly with the number of visual tokens.
arXiv:2505.16157v3 Announce Type: replace Abstract: Transformer-based models have made remarkable progress in image restoration (IR) tasks. However, the quadratic complexity of self-attention in Tran...
The paper introduces Semantic Head Specialization (SHS), a phenomenon where Vision Transformer (ViT) attention heads specialize as either object- or background-focused, most evident under full attention. It proposes the SHS-Index to quantify this specialization, demonstrating its ability to distinguish full-attention from chunk-window ViTs and its strong correlation with downstream benchmark performance. Leveraging insights into window interaction, token serialization, and local softmax allocation, the authors design Ariadne Attention, a hybrid attention mechanism that matches full-attention performance on 22 image and video tasks while reducing attention compute by 6.5×.
arXiv:2506. 08297v2 Announce Type: replace-cross Abstract: Attention is the critical component of a transformer.
arXiv:2509. 06461v3 Announce Type: replace-cross Abstract: Vision-Language Models (VLMs) have demonstrated remarkable success across diverse visual tasks, yet their performance degrades in complex visual environments.
Video DeltaNet (VDN) introduces a hybrid attention mechanism for video diffusion models, combining local Softmax attention with a bidirectional linear memory branch called Video Delta Attention (VDA). The design updates memory once per frame, uses separate output projections and learnable gates to balance the two branches, and employs a staged teacher‑alignment recipe to integrate the new pathway into pretrained models. When applied to MiniMax H3, VDN achieves a 14.5× speedup, completing 14.3‑second, 768p video denoising in 6.70 seconds on eight NVIDIA B200 GPUs compared to the 50‑step dense baseline.
Diffusion Transformers (DiTs) incur quadratic self-attention cost over spatiotemporal tokens. Existing training-free sparse attention methods often construct sparse masks from block-level or cluster-level proxy scores, which can obscure fine-grained differences among keys and miss high contribution keys under aggressive sparsity.
Crane is a CLIP‑based framework for zero‑shot anomaly detection that enhances dense localization by adapting the vision encoder with a correlation‑based attention module and conditioning learnable prompts on global image context. It further fuses anomaly‑relevant patch features into the global representation for more sensitive image‑level detection, and a variant called Crane+ leverages DINOv2 spatial correlations for stronger pixel‑level performance. Across seven industrial benchmarks, Crane raises mean image‑level AP by 4.5% and Crane+ boosts mean pixel‑level AUPRO by 9.0%.
The paper introduces QK Product Steering, a data‑free, training‑free method that edits the query‑key product in vision‑language models to reduce object hallucination. By suppressing a few dominant singular modes in selected middle layers and mapping the edited product back to query weights, the approach lowers hallucination rates without affecting inference cost. Experiments on three GQA‑based VLMs show a 4.0% average reduction in CHAIR$_s$, with the effect localized to symmetric mutual‑attention channels.
arXiv:2604. 00086v2 Announce Type: replace-cross Abstract: The field of computer vision has experienced significant advancements through scalable vision encoders and multimodal pre-training frameworks.
The paper introduces STD, a hierarchical token pruning framework for Large Vision‑Language Models that aligns pruning strategies with the functional roles of different network stages. By using high‑frequency spectral analysis in shallow layers, Gaussian‑smoothed attention in intermediate layers, and a stability‑adaptive trigger in deep layers, STD preserves essential visual information while aggressively reducing token counts. Experiments demonstrate that STD outperforms existing pruning methods, achieving up to 94.4% token reduction and a 3.9× speed‑up on LLaVA‑NeXT‑7B.
arXiv:2607.09086v2 Announce Type: replace Abstract: We present Subtoken Vision Transformer (SubViT), a selective image tokenization method for fine-grained visual recognition. Standard Vision Transfo...