ScreenSpot-Pro

The 'ScreenSpot-pro' dataset/benchmark contains challenging data for evaluating the accuracy of end-to-end grounding models in graphical user interface (GUI) tasks, specifically focusing on actions like clicking and typing.

Papers using ScreenSpot-Pro (10)

Mitigating Coordinate Prediction Bias From Positional Encoding Failures2026Kimi-VL Technical Report2025 · 1 citesMEGA-GUI: Multi-stage Enhanced Grounding Agents For GUI Elements2025Adaptive Vision-Language Model Routing for Computer Use Agents2026MolmoPoint: Better Pointing for VLMs with Grounding Tokens2026GUI-AIMA: Aligning Intrinsic Multimodal Attention with a Context Anchor for GUI Grounding2025\textsc{gui-spotlight}: Adaptive Iterative Focus Refinement For Enhanced GUI Visual Grounding2025Phi-ground Tech Report: Advancing Perception In GUI Grounding2025Gui-shift: Enhancing Vlm-based GUI Agents Through Self-supervised Reinforcement Learning2025Chain-of-ground: Improving GUI Grounding Via Iterative Reasoning And Reference Feedback2025