Sparse Attention for Dense Open-Vocabulary Prediction in CLIP

• URL: http://arxiv.org/abs/2607.07135v2
• Date: Mon, 13 Jul 2026 05:58:12 GMT
• Title: Sparse Attention for Dense Open-Vocabulary Prediction in CLIP
• Abstract summary: We study an inference-time substitution of the row-wise softmax in the final visual self-attention layers.<n>Because entmax applies a data-dependent threshold that maps low scores exactly to zero, it acts as an implicit denoiser.<n>We evaluate on open-vocabulary tasks-dense semantic segmentation (Pascal VOC, Pascal Context, ADE20K) and fine-grained retrieval (FG-OVD)
• Score: 54.45759517856271
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Contrastive Language-Image Pre-training (CLIP) relies on softmax-based self-attention, a strictly positive distribution that assigns probability mass to every pair of tokens-even semantically irrelevant ones. While these dense softmax weights are effective for gathering broad context during pre-training, they spread attention across many low-salience tokens, producing noise that obscures the fine-grained, spatially localized cues required for dense, open-vocabulary prediction. We study an inference-time substitution of the row-wise softmax in the final visual self-attention layers with the $α$-entmax transform, applied across both the standard query-key attention and self-correlation variants. Because entmax applies a data-dependent threshold that maps low scores exactly to zero, it acts as an implicit denoiser, zeroing contextually irrelevant dependencies while redistributing mass onto the most relevant tokens. We evaluate on open-vocabulary tasks-dense semantic segmentation (Pascal VOC, Pascal Context, ADE20K) and fine-grained retrieval (FG-OVD)-and find the gain from attention sparsification is proportional to how much the baseline attention spreads off the target class.

Related papers

• Attention-Discounted Adaptive Sampler for Masked Diffusion Language Models [59.51249894128724]
  Masked diffusion language models can reduce inference steps by revealing multiple tokens per denoising iteration.<n>We propose ADAS, a training-free reranking rule for parallel masked diffusion decoding.
  arXiv  Detail & Related papers  (2026-06-09T13:17:27Z)
• The Silent Vote: Improving Zero-Shot LLM Reliability by Aggregating Semantic Neighborhoods [0.0]
  We propose an inference-time layer that recovers lost information by aggregating the scores of the semantic neighborhood surrounding each target label.<n>Our results demonstrate consistent improvements across all evaluation metrics: Expected Error (ECE) and Brier Score, while simultaneously enhancing discriminative performance in terms of AUROC and Macro-F1.
  arXiv  Detail & Related papers  (2026-05-10T20:22:10Z)
• Evading Visual Aphasia: Contrastive Adaptive Semantic Token Pruning for Vision-Language Models [52.78477729846771]
  We introduce COAST (COntrastive Adaptive Semantic Token Pruning), a training-free pruning framework that casts compression as adaptive semantic routing.<n> COAST consistently outperforms strong pruning baselines across token budgets and generalizes across multiple LVLM families.
  arXiv  Detail & Related papers  (2026-05-10T09:07:04Z)
• ZScribbleSeg: A comprehensive segmentation framework with modeling of efficient annotation and maximization of scribble supervision [15.414625894259844]
  We investigate the principle of good scribble annotations, which leads to efficient randomness forms via supervision and simulation.<n>We introduce regularization terms to encode the spatial relationship and the shape constraints, where the EM algorithm is utilized to estimate the mixture ratios of label classes.<n>We integrate the efficient supervision with the prior into a framework, referred to as ZScribbleSeg, and apply it to multiple scenarios.
  arXiv  Detail & Related papers  (2026-05-07T13:41:15Z)
• Learning by Neighbor-Aware Semantics, Deciding by Open-form Flows: Towards Robust Zero-Shot Skeleton Action Recognition [41.77490816513839]
  We propose a novel method for zero-shot skeleton action recognition, termed $texttt$textbfFlora$$.<n>Specifically, we attune textual semantics by incorporating direction-aware regional semantics, and a cross-modal consistency objective.<n>Experiments on three benchmark datasets validate the effectiveness of our method, showing particularly impressive performance even when trained with only 10% of the seen data.
  arXiv  Detail & Related papers  (2025-11-12T14:54:53Z)
• Long-Context Generalization with Sparse Attention [21.400056571592277]
  Transformer-based architectures traditionally employ softmax to compute attention weights.<n>As sequence length increases, non-informative tokens accumulate attention probability mass, leading to dispersion and representational collapse.<n>We show that dynamically sparse attention mechanisms using $alpha$-entmax can avoid these issues, due to their ability to assign exact zeros to irrelevant tokens.
  arXiv  Detail & Related papers  (2025-06-19T22:43:25Z)
• HYPEROFA: Expanding LLM Vocabulary to New Languages via Hypernetwork-Based Embedding Initialization [50.27950279695363]
  Many pre-trained language models (PLMs) exhibit suboptimal performance on mid- and low-resource languages.<n>A common strategy to address this is to introduce new tokens specific to the target languages, initialize their embeddings, and apply continual pre-training on target-language data.<n>We propose HYPEROFA, a hypernetwork-based approach for more adaptive token embedding.
  arXiv  Detail & Related papers  (2025-04-21T19:40:32Z)
• Semantic Equitable Clustering: A Simple and Effective Strategy for Clustering Vision Tokens [57.37893387775829]
  We introduce a fast and balanced clustering method, named Semantic Equitable Clustering (SEC)<n>SEC clusters tokens based on their global semantic relevance in an efficient, straightforward manner.<n>We propose a versatile vision backbone, SECViT, to serve as a vision language connector.
  arXiv  Detail & Related papers  (2024-05-22T04:49:00Z)
• ZScribbleSeg: Zen and the Art of Scribble Supervised Medical Image Segmentation [16.188681108101196]
  We propose to utilize solely scribble annotations for weakly supervised segmentation. Existing solutions mainly leverage selective losses computed solely on annotated areas. We introduce regularization terms to encode the spatial relationship and shape prior. We integrate the efficient scribble supervision with the prior into a unified framework, denoted as ZScribbleSeg.
  arXiv  Detail & Related papers  (2023-01-12T09:00:40Z)
• DenseCLIP: Extract Free Dense Labels from CLIP [130.3830819077699]
  Contrastive Language-Image Pre-training (CLIP) has made a remarkable breakthrough in open-vocabulary zero-shot image recognition. DenseCLIP+ surpasses SOTA transductive zero-shot semantic segmentation methods by large margins. Our finding suggests that DenseCLIP can serve as a new reliable source of supervision for dense prediction tasks.
  arXiv  Detail & Related papers  (2021-12-02T09:23:01Z)
• PCPL: Predicate-Correlation Perception Learning for Unbiased Scene Graph Generation [58.98802062945709]
  We propose a novel Predicate-Correlation Perception Learning scheme to adaptively seek out appropriate loss weights. Our PCPL framework is further equipped with a graph encoder module to better extract context features.
  arXiv  Detail & Related papers  (2020-09-02T08:30:09Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.