Built independently by an author, for readers. Read the story and support ChapterPal
ChapterPal Logo
Log inSign up

keyword

multi-task visual grounding

Multi-task visual grounding is a multimodal artificial intelligence paradigm that unifies multiple forms of language-guided visual localization within a single computational framework. Given an input image and a natural language query, the system simultaneously predicts different spatial representations of the referenced target, primarily integrating bounding-box localization for referring expression comprehension with pixel-level mask generation for referring expression segmentation. By sharing representations across these complementary objectives and extending across multiple semantic granularities—including instance, semantic, and part levels—multi-task visual grounding allows models to improve cross-modal alignment, resolve visual ambiguities, and maintain spatial and semantic consistency across diverse vision-language localization outputs.

1 item

Hierarchical Open-vocabulary Universal Image Segmentation

Hierarchical Open-vocabulary Universal Image Segmentation

Xudong Wang, Shufan Li, Konstantinos Kallidromitis, Yusuke Kato, Kazuki Kozuka, Trevor Darrell

OrganizationsPanasonicUniversity of California Berkeley

Why you should read this

Presents HIPIE, a unified open-vocabulary framework that resolves segmentation ambiguity across multiple granularities by incorporating hierarchical visual representations alongside decoupled text-image fusion mechanisms for stuff and thing categories.

Open-vocabulary image segmentation aims to partition an image into semantic regions according to arbitrary text descriptions. However, complex visual scenes can be naturally decomposed into simpler parts and abstracted at multiple levels of granularity, introducing inherent segmentation ambiguity. Unlike existing methods that typically sidestep this ambiguity and treat it as an external factor, our approach actively incorporates a hierarchical representation encompassing different semantic-levels into the learning process. We also propose a decoupled text-image fusion mechanism and representation learning modules for both “things” and “stuff”.1 Additionally, we systematically examine the differences that exist in the textual and visual features between these types of categories. Our resulting model, named HIPIE, tackles HIerarchical, oPen-vocabulary, and unIvErsal segmentation tasks within a unified framework. Benchmarked on over 40 datasets, e.g., ADE20K, COCO, Pascal-VOC Part, RefCOCO/RefCOCOg, ODinW and SeginW, HIPIE achieves the state-of-the-art results at various levels of image comprehension, including semantic-level (e.g., semantic segmentation), instance-level (e.g., panoptic/referring segmentation and object detection), as well as part-level (e.g., part/subpart segmentation) tasks.

Added

2026-09-26