Research map / 3D vision / 3D generation
MVP: Multiple View Prediction Improves GUI Grounding
Yunzhu Zhang, Zeyu Pan, Zhengwen Zeng, Shuheng Shen, Changhua Meng, Linchao Zhu
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 attention-guided views instead of border padding (ablation) | MVP · ScreenSpot-Pro (GTA1-7B ablation, 4 views) | accuracy (%): 57.3 -> 61.7 | Improved |
| add border padding multi-view inference over single image (ablation) | MVP baseline · ScreenSpot-Pro (GTA1-7B ablation, 4 views) | accuracy (%): 49.8 -> 57.3 | Improved |
| use clustering instead of averaging coordinates (ablation) | MVP · ScreenSpot-Pro (GTA1-7B ablation, 4 views) | accuracy (%): 46.6 -> 61.7 | Improved |
| use clustering instead of random selection (ablation) | MVP · ScreenSpot-Pro (GTA1-7B ablation, 4 views) | accuracy (%): 55.7 -> 61.7 | Improved |
| add 2x view resizing (ablation) | MVP · ScreenSpot-Pro (GTA1-7B ablation, 4 views) | accuracy (%): 59.1 -> 61.7 | Improved |
| use comma token as attention query instead of instruction tokens (ablation) | MVP · ScreenSpot-Pro (GTA1-7B ablation, 4 views) | accuracy (%): 60.5 -> 61.7 | Improved |
| use comma token instead of im_end token as query (ablation) | MVP · ScreenSpot-Pro (GTA1-7B ablation, 4 views) | accuracy (%): 33.3 -> 61.7 | Improved |
| increase number of views beyond 4 (ablation) | MVP · ScreenSpot-Pro (GTA1-7B ablation, 4 views) | accuracy (%) | No clear effect |
| replace clustering with trained selector (GTA1-7B) (ablation) | MVP · ScreenSpot-Pro (GTA1-7B ablation, 4 views) | accuracy (%): 61.7 -> 62.8 | Mixed |
| replace clustering with trained selector (Qwen3VL-8B) (ablation) | MVP · ScreenSpot-Pro (Qwen3VL-8B-Instruct, selector ablation) | accuracy (%): 65.5 -> 65.3 | No clear effect |
| add MVP to UI-TARS-1.5-7B | UI-TARS-1.5-7B · ScreenSpot-Pro | accuracy (%): 41.9 -> 56.1 | Improved |
| add MVP to Qwen3VL-32B-Instruct | Qwen3VL-32B-Instruct · ScreenSpot-Pro | accuracy (%): 55.3 -> 74.0 | Improved |
Closest papers by meaning
- FocusUI: Efficient UI Grounding via Position-Preserving Visual Token Selection (CVPR 2026 · similarity 0.90)
- GUI-Spotlight: Adaptive Iterative Focus Refinement for Enhanced GUI Visual Grounding (ICML 2026 · similarity 0.89)
- Trifuse: Enhancing Attention-Based GUI Grounding via Multimodal Fusion (ICML 2026 · similarity 0.89)
- GUI-Actor: Coordinate-Free Visual Grounding for GUI Agents (NeurIPS 2025 · similarity 0.89)
- Towards GUI Agents: Vision-Language Diffusion Models for GUI Grounding (CVPR 2026 · similarity 0.88)
- BAMI: Training-Free Bias Mitigation in GUI Grounding (CVPR 2026 · 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.
