Research map / Vision-language and multimodal / Robot action and navigation
Visual Test-time Scaling for GUI Agent Grounding
Tiange Luo, Lajanugen Logeswaran, Justin Johnson, Honglak Lee
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add RegionFocus to UI-TARS-72B | UI-TARS-72B · ScreenSpot-Pro | average accuracy: 38.1 -> 50.2 | Improved |
| add RegionFocus to Qwen2.5-VL-72B | Qwen2.5-VL-72B · ScreenSpot-Pro | average accuracy: 47.8 -> 61.6 | Improved |
| add RegionFocus to UI-TARS-7B | UI-TARS-7B · ScreenSpot-Pro | average accuracy: 35.7 -> 41.2 | Improved |
| add RegionFocus to Qwen2.5-VL-7B | Qwen2.5-VL-7B · ScreenSpot-Pro | average accuracy: 26.8 -> 32.1 | Improved |
| add RegionFocus to UI-TARS-72B | UI-TARS-72B · WebVoyager | overall success rate: 44.1% -> 59.5% | Improved |
| add RegionFocus to Qwen2.5-VL-72B | Qwen2.5-VL-72B · WebVoyager | overall success rate: 42.4% -> 52.7% | Improved |
| add RegionFocus to UI-TARS-7B | UI-TARS-7B · WebVoyager | overall success rate: 33.2% -> 44.7% | Improved |
| add RegionFocus to Qwen2.5-VL-7B | Qwen2.5-VL-7B · WebVoyager | overall success rate: 32.5% -> 41.1% | Improved |
| add RegionFocus to Qwen2.5-VL-7B on individual sites | Qwen2.5-VL-7B · WebVoyager (Google Map) | success rate: 30.2% -> 17.1% | Hurt |
| use image-as-map instead of text-as-history for region history and aggregation (ablation) | RegionFocus (UI-TARS-7B-DPO) · WebVoyager subset | success rate: 37.2 -> 43.2 | Improved |
| derive fixed bounding boxes from predicted focal point instead of predicting regions directly (ablation) | RegionFocus (UI-TARS-7B-DPO) · WebVoyager subset | success rate: 28.1 -> 43.2 | Improved |
| replace fixed boxes with SAM-generated boxes (ablation) | RegionFocus (UI-TARS-7B-DPO) · WebVoyager subset | success rate: 43.2 -> 46.5 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- OS-Atlas-7B — OS-ATLAS: Foundation Action Model for Generalist GUI Agents (ICLR 2025) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- GTA1: GUI Test-time Scaling Agent (ICLR 2026 · similarity 0.90)
- GUI-Actor: Coordinate-Free Visual Grounding for GUI Agents (NeurIPS 2025 · similarity 0.90)
- Navigating the Digital World as Humans Do: Universal Visual Grounding for GUI Agents (ICLR 2025 · similarity 0.89)
- FocusUI: Efficient UI Grounding via Position-Preserving Visual Token Selection (CVPR 2026 · similarity 0.88)
- GUI-Spotlight: Adaptive Iterative Focus Refinement for Enhanced GUI Visual Grounding (ICML 2026 · similarity 0.88)
- ShowUI: One Vision-Language-Action Model for GUI Visual Agent (CVPR 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
