ScreenHaystack: Finding Blind Zones in GUI Grounding
Chenyue Li , Xiaoxiao Sun , Yubo Deng , Qinlin Zhao , Serena Yeung-Levy , Yuhui Zhang
- 🏛 Institutions
- Stanford
- 📅 Date
- September 25, 2026
- 📑 Publisher
- EMNLP 2026
- 💻 Env
- General GUI
- 🔑 Keywords
TLDR
ScreenHaystack is a needle-in-a-haystack benchmark that relocates controlled target icons across high-resolution GUI backgrounds, revealing blind zones where grounding accuracy drops sharply for models including Qwen3-VL, UI-TARS, GTA, and UI-Venus. The blind zones transfer to ScreenSpot-Pro, and augmenting training data in them improves accuracy.
Related papers (24)
- IGG: A Benchmark for Interactive GUI Grounding under Visibility ConstraintsOctober 23, 2026 · arXiv
- PointerBench: A Diagnostic Benchmark for Fine-Grained GUI GroundingOctober 22, 2026 · arXiv
- How Many Pixels Is a Digit Worth? Place-Aware Coordinate Entropy for GUI Agent Confidence EstimationSeptember 21, 2026 · EMNLP 2026
- Learning from Reliable Negatives: Confidence-Anchored Test-Time Adaptation for GUI GroundingSeptember 14, 2026 · arXiv
- GUI-Perturbed: Domain Randomization Reveals Systematic Brittleness in GUI Grounding ModelsApril 15, 2026 · arXiv
- What's Missing in Screen-to-Action? Towards a UI-in-the-Loop Paradigm for Multimodal GUI ReasoningApril 8, 2026 · Findings of ACL 2026
- Beyond Clicking: A Step Towards Generalist GUI Grounding via Text DraggingNovember 7, 2025 · arXiv
- GUI-AIMA: Aligning Intrinsic Multimodal Attention with a Context Anchor for GUI GroundingNovember 2, 2025 · ECCV 2026
- Scaling Computer‑Use Grounding via User Interface Decomposition and SynthesisMay 19, 2025 · NeurIPS 2025 Datasets and Benchmarks Track (Spotlight)
- UI-E2I-Synth: Advancing GUI Grounding with Large-Scale Instruction SynthesisApril 15, 2025 · Findings of ACL 2025
- ConClick: Element-Wise Contrastive Learning Improves GUI GroundingOctober 9, 2026 · arXiv
- VenusBench-GD: A Comprehensive Multi-Platform GUI Benchmark for Diverse Grounding TasksDecember 18, 2025 · arXiv
- ScreenSpot-Pro: GUI Grounding for Professional High-Resolution Computer UseApril 4, 2025 · ACM Multimedia 2025
- SeeClick: Harnessing GUI Grounding for Advanced Visual GUI AgentsJanuary 17, 2024 · ACL 2024
- Right Click, Wrong Box: Grading GUI Grounding by What a Click DoesOctober 27, 2026 · arXiv
- Learning to Look Beyond Screens: Transferable Visual Observation from GUI GroundingOctober 27, 2026 · arXiv
- AdaGround: Token-Normalized Adaptive Zooming with Verifier-in-the-Loop Refinement for GUI GroundingOctober 27, 2026 · arXiv
- PreView: Enhancing GUI Grounding with Preprocessed Views and CuesOctober 24, 2026 · arXiv
- Beyond Next-Screen Prediction: The Decision Value of GUI World ModelsOctober 17, 2026 · arXiv
- MMA-SafetyBench: A Benchmark for Multimodal Agent Safety EvaluationOctober 14, 2026 · arXiv
- GroundGUI: GUI Grounding by High-Quality Data and Distillation-to-Reinforcement LearningOctober 13, 2026 · arXiv
- cua-speedrun: Standardized Benchmarking of the Speed of Computer-Use AgentsSeptember 30, 2026 · arXiv
- SCA: Spatial Credit Assignment for Reinforcement Learning of GUI AgentsSeptember 29, 2026 · arXiv
- PrecogUI: Proactive GUI Agents via Pre-cognitive Simulation and Experience RetrievalSeptember 29, 2026 · arXiv