Open-Vocabulary Semantic Segmentation

Overview

Open-vocabulary semantic segmentation aims to assign meaningful labels to each pixel, including categories not seen during training, making it useful when the set of objects or environments can change. Many methods reuse vision-language models such as CLIP, DINO, and SAM, combining text–image similarity with attention, mask proposals, feature alignment, or prompt and prototype calibration; others train with region-level text, pseudo-labels, or generated data. Systems are tested on general benchmarks such as Pascal, COCO, and ADE20K, as well as remote-sensing, urban-scene, 3D, and other specialized settings. A central challenge is balancing broad category recognition with precise boundaries and small-object localization, especially under domain shifts, while keeping inference efficient and evaluation protocols comparable.

Latest papers

CardsList

93 papers6 papers in the last four weeks, down 14% on the four weeks before. 0.0% of all new papers.

  1. SynSeg: Feature Synergy for Multi-Category Contrastive Learning in End-to-End Open-Vocabulary Semantic Segmentation

    Aug 8, 2025Weichen Zhang, Kebin Liu, Fan Dang +3Contrastive LearningWeakly Supervised Image Segmentation

  2. LlamaSeg: Image Segmentation via Autoregressive Mask Generation

    May 26, 2025Jiru Deng, Tengjin Weng, Tianyu Yang +3Image SegmentationText-Guided Image Segmentation

  3. Synergistic Vision-Language Reinforcement Enables Scalable On-Demand Analysis across Diverse Clinical Tasks

    May 6, 2025Haonan Wang, Jiaji Mao, Lehan Wang +11Image SegmentationMedical Imaging Foundation Models