Summary

This repository serves as a comprehensive, community-driven survey and curated collection of research papers, code repositories, and project resources focused on Multimodal Referring Segmentation. It addresses the challenge of systematically organizing the rapidly expanding literature in this domain, providing researchers and practitioners with a centralized, navigable overview. The primary benefit is facilitating efficient discovery and evaluation of state-of-the-art methods across diverse referring segmentation tasks, from 2D images and videos to 3D scenes and audio-visual contexts.

How It Works

The project functions as a living document, meticulously cataloging and categorizing a wide array of research contributions. It is structured into distinct sections covering Referring Expression Segmentation (RES), Referring Video-Object Segmentation (RVOS), Referring Audio-Visual Segmentation (RAVS), 3D Referring Expression Segmentation (3D-RES), Generalized Referring Expression x (GREx), and specific applications. Each entry typically includes the paper title, source (e.g., arXiv link), and crucially, direct links to associated code repositories or project homepages, enabling practical evaluation and adoption.

Highlighted Details

• Extensive and categorized listing of research papers, covering a broad spectrum from foundational work to recent advancements in multimodal referring segmentation.
• Emphasis on practical implementation, with a high proportion of entries linking directly to publicly available code (e.g., GitHub repositories), facilitating reproducibility and adoption.
• Active community engagement model, encouraging contributions to ensure the survey remains comprehensive and up-to-date.
• Coverage spans multiple modalities and dimensions, including text, image, video, audio, and 3D point clouds/scenes.

Maintenance & Community

The repository actively solicits community contributions via pull requests, inviting users to suggest omitted works, implementations, or resources. This collaborative approach aims to maintain the survey's comprehensiveness and relevance in a fast-paced research field. Links to community platforms like Discord or Slack are not explicitly mentioned.

Licensing & Compatibility

No specific software license is detailed in the provided README content. Users should verify the licenses of individual projects linked within the survey for compatibility with their intended use cases, especially for commercial applications.

Limitations & Caveats

The survey acknowledges the inherent challenge of exhaustively covering all published research, particularly given the high volume of preprints on platforms like arXiv. While striving for broad inclusion, some recent or highly specialized works may not yet be listed. The focus is on curating existing research rather than providing a runnable codebase itself.