From Web to Pixels: Bringing Agentic Search into Visual Perception (WebEye & Pixel-Searcher)

Forum topic · 小凯 · 2026-05-14

Summary

This arXiv paper (2605.12497) introduces Perception Deep Research, an open-world visual perception challenge where a visible object must first be resolved from external facts, recent events, long-tail entities, or multi-hop relations before it can be localized. The authors present WebEye, an object-anchored benchmark featuring verifiable evidence, knowledge-intensive queries, and precise box/mask annotations across three task views: Search-based Grounding, Search-based Segmentation, and Search-based VQA. WebEye contains 120 images, 473 annotated object instances, 645 unique QA pairs, and 1927 task samples. They also propose Pixel-Searcher, an agentic search-to-pixel workflow that resolves hidden target identities and binds them to boxes, masks, or grounded answers. Experiments show Pixel-Searcher achieves the strongest open-source performance across all three tasks, with failures mainly stemming from evidence acquisition, identity resolution, and visual instance binding.

论文概要

研究领域: CV (Computer Vision) Authors: Bokang Yang, Xinyi Sun, Kaituo Feng, Xingping Dong, Dongming Wu, Xiangyu Yue Published: 2026-05-12 arXiv: 2605.12497

Abstract (translated)

Visual perception connects high-level semantic understanding to pixel-level perception, but most existing settings assume that the decisive evidence for identifying a target is already in the image or frozen model knowledge. We study a more practical yet harder open-world case where a visible object must first be resolved from external facts, recent events, long-tail entities, or multi-hop relations before it can be localized. We formalize this challenge as Perception Deep Research and introduce WebEye, an object-anchored benchmark with verifiable evidence, knowledge-intensive queries, precise box/mask annotations, and three task views: Search-based Grounding, Search-based Segmentation, and Search-based VQA. WebEye contains 120 images, 473 annotated object instances, 645 unique QA pairs, and 1927 task samples. We further propose Pixel-Searcher, an agentic search-to-pixel workflow that resolves hidden target identities and binds them to boxes, masks, or grounded answers. Experiments show that Pixel-Searcher achieves the strongest open-source performance across all three task views, with failures mainly arising from evidence acquisition, identity resolution, and visual instance binding.

Key Resources

arXiv: https://arxiv.org/abs/2605.12497

---

*Auto-collected on 2026-05-14*