Learning Sparse Visual Representations via Spatial-Semantic Factorization

• URL: http://arxiv.org/abs/2602.01905v1
• Date: Mon, 02 Feb 2026 10:12:17 GMT
• Title: Learning Sparse Visual Representations via Spatial-Semantic Factorization
• Authors: Theodore Zhengde Zhao, Sid Kiblawi, Jianwei Yang, Naoto Usuyama, Reuben Tan, Noel C Codella, Tristan Naumann, Hoifung Poon, Mu Wei,
• Abstract summary: Self-supervised learning (SSL) faces a fundamental conflict between semantic understanding and image reconstruction.<n>We introduce STELLAR, a framework that factorizes visual features into a low-rank product of semantic concepts and their spatial distributions.<n>We demonstrate that as few as 16 sparse tokens under this factorized form are sufficient to simultaneously support high-quality reconstruction (2.60 FID) and match the semantic performance of dense backbones (79.10% ImageNet accuracy)
• Score: 37.169502692169196
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Self-supervised learning (SSL) faces a fundamental conflict between semantic understanding and image reconstruction. High-level semantic SSL (e.g., DINO) relies on global tokens that are forced to be location-invariant for augmentation alignment, a process that inherently discards the spatial coordinates required for reconstruction. Conversely, generative SSL (e.g., MAE) preserves dense feature grids for reconstruction but fails to produce high-level abstractions. We introduce STELLAR, a framework that resolves this tension by factorizing visual features into a low-rank product of semantic concepts and their spatial distributions. This disentanglement allows us to perform DINO-style augmentation alignment on the semantic tokens while maintaining the precise spatial mapping in the localization matrix necessary for pixel-level reconstruction. We demonstrate that as few as 16 sparse tokens under this factorized form are sufficient to simultaneously support high-quality reconstruction (2.60 FID) and match the semantic performance of dense backbones (79.10% ImageNet accuracy). Our results highlight STELLAR as a versatile sparse representation that bridges the gap between discriminative and generative vision by strategically separating semantic identity from spatial geometry. Code available at https://aka.ms/stellar.

Related papers

• Modality Gap-Driven Subspace Alignment Training Paradigm For Multimodal Large Language Models [84.78794648147608]
  A persistent geometric anomaly, the Modality Gap, remains.<n>Prior approaches to bridge this gap are largely limited by oversimplified isotropic assumptions.<n>We propose the Fixed-frame Modality Gap Theory, which decomposes the modality gap into stable biases and anisotropic residuals.<n>We then introduce ReAlign, a training-free modality alignment strategy.
  arXiv  Detail & Related papers  (2026-02-02T13:59:39Z)
• Both Semantics and Reconstruction Matter: Making Representation Encoders Ready for Text-to-Image Generation and Editing [62.94394079771687]
  A burgeoning trend is to adopt high-dimensional features from representation encoders as generative latents.<n>We propose a systematic framework to adapt understanding-oriented encoder features for generative tasks.<n>We show that our approach achieves state-of-the-art reconstruction, faster convergence, and substantial performance gains in both Text-to-Image (T2I) and image editing tasks.
  arXiv  Detail & Related papers  (2025-12-19T18:59:57Z)
• Adapting Self-Supervised Representations as a Latent Space for Efficient Generation [18.746963205066688]
  RepTok is a generative modeling framework that represents an image using a single continuous latent token.<n>RepTok achieves competitive results on class-conditional ImageNet generation and naturally extends to text-to-image synthesis.
  arXiv  Detail & Related papers  (2025-10-16T12:43:03Z)
• CoPatch: Zero-Shot Referring Image Segmentation by Leveraging Untapped Spatial Knowledge in CLIP [26.827036116024914]
  textscCoPatch is a zero-shot RIS framework that enhances spatial representations in both text and image modalities.<n>We show that textscCoPatch significantly improves spatial grounding in zero-shot RIS across RefCOCO, RefCOCO+, RefCOCOg, and PhraseCut (+ 2--7 mIoU) without requiring any additional training.
  arXiv  Detail & Related papers  (2025-09-27T04:12:10Z)
• Locality Alignment Improves Vision-Language Models [55.275235524659905]
  Vision language models (VLMs) have seen growing adoption in recent years, but many still struggle with basic spatial reasoning errors.<n>Our goal is to resolve this with a vision backbone that effectively captures both local and global image semantics.<n>We propose a new efficient post-training stage for ViTs called locality alignment and a novel fine-tuning procedure called MaskEmbed.
  arXiv  Detail & Related papers  (2024-10-14T21:01:01Z)
• Spatial Semantic Recurrent Mining for Referring Image Segmentation [63.34997546393106]
  We propose Stextsuperscript2RM to achieve high-quality cross-modality fusion. It follows a working strategy of trilogy: distributing language feature, spatial semantic recurrent coparsing, and parsed-semantic balancing. Our proposed method performs favorably against other state-of-the-art algorithms.
  arXiv  Detail & Related papers  (2024-05-15T00:17:48Z)
• Spatial Latent Representations in Generative Adversarial Networks for Image Generation [0.0]
  We define a family of spatial latent spaces for StyleGAN2. We show that our spaces are effective for image manipulation and encode semantic information well.
  arXiv  Detail & Related papers  (2023-03-25T20:01:11Z)
• Learning Aligned Cross-Modal Representation for Generalized Zero-Shot Classification [17.177622259867515]
  We propose an innovative autoencoder network by learning Aligned Cross-Modal Representations (dubbed ACMR) for Generalized Zero-Shot Classification (GZSC) Specifically, we propose a novel Vision-Semantic Alignment (VSA) method to strengthen the alignment of cross-modal latent features on the latent subspaces guided by a learned classifier. In addition, we propose a novel Information Enhancement Module (IEM) to reduce the possibility of latent variables collapse meanwhile encouraging the discriminative ability of latent variables.
  arXiv  Detail & Related papers  (2021-12-24T03:35:37Z)
• HSVA: Hierarchical Semantic-Visual Adaptation for Zero-Shot Learning [74.76431541169342]
  Zero-shot learning (ZSL) tackles the unseen class recognition problem, transferring semantic knowledge from seen classes to unseen ones. We propose a novel hierarchical semantic-visual adaptation (HSVA) framework to align semantic and visual domains. Experiments on four benchmark datasets demonstrate HSVA achieves superior performance on both conventional and generalized ZSL.
  arXiv  Detail & Related papers  (2021-09-30T14:27:50Z)

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.