Research map / Vision-language and multimodal / Multimodal LLMs
GUI-AIMA: Aligning Intrinsic Multimodal Attention with a Context Anchor for GUI Grounding
Shijie Zhou, Viet Lai, Hao Tan, Jihyung Kil, Wanrong Zhu, Changyou Chen, Ruiyi Zhang
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 ANCHOR-token aggregation with uniform head weights (simplified attention grounding) over vanilla token-wise attention (ablation) | GUI-AIMA (45k) · ScreenSpot-Pro | accuracy: 37.51 -> 40.23 | Improved |
| weight heads with visual-sink top-1 query token (global) instead of uniform weights (ablation) | GUI-AIMA (45k) · ScreenSpot-Pro | accuracy: 40.23 -> 42.13 | Improved |
| weight heads with all query tokens (TAG style) instead of uniform (ablation) | GUI-AIMA (45k) · ScreenSpot-Pro | accuracy: 40.23 -> 37.63 | Hurt |
| weight heads with ANCHOR token only instead of uniform (ablation) | GUI-AIMA (45k) · ScreenSpot-v2 | accuracy: 88.92 -> 87.81 | Hurt |
| use top-3 instead of top-1 global visual-sink tokens (ablation) | GUI-AIMA (45k) · ScreenSpot-v2 | accuracy: 89.39 -> 84.98 | Hurt |
| use layer-wise instead of global shared visual-sink token (ablation) | GUI-AIMA (45k) · ScreenSpot-v2 / ScreenSpot-Pro | accuracy: SSv2 89.39 -> 89.86; SSPro 42.13 -> 41.43 | Mixed |
| add overlap- and distance-aware weighted patch labeling (ablation) | GUI-AIMA (45k) · ScreenSpot-Pro | accuracy: 42.13 -> 43.39 | Improved |
| add two-step zoom-in inference | GUI-AIMA-3B · ScreenSpot-Pro | accuracy: 53.8 -> 61.5 | Improved |
| add zoom-in inference | GUI-AIMA-3B · OSWorld-G-Refine | accuracy: 62.8 -> 68.1 | Improved |
| crop and zoom second step by 2x versus crop-only (ablation) | GUI-AIMA two-step inference · ScreenSpot-Pro offset errors | improved predictions: 49 -> 182 | Improved |
| apply GUI-AIMA to InternVL3.5-4B backbone (ablation) | InternVL3.5-4B · ScreenSpot-v2 | accuracy: 85.1 -> 88.8 | Improved |
| stop gradient on head weights (ablation) | GUI-AIMA-3B-lite (254k) · ScreenSpot-Pro | accuracy: 49.78 -> 45.73 | Hurt |
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.
- GUI-Actor-3B — GUI-Actor: Coordinate-Free Visual Grounding for GUI Agents (NeurIPS 2025) · better on 3 of 3 shared results
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
- Trifuse: Enhancing Attention-Based GUI Grounding via Multimodal Fusion (ICML 2026 · similarity 0.92)
- GUI-Spotlight: Adaptive Iterative Focus Refinement for Enhanced GUI Visual Grounding (ICML 2026 · similarity 0.92)
- Attention-Driven GUI Grounding: Leveraging Pretrained Multimodal Large Language Models Without Fine-Tuning (AAAI 2025 · similarity 0.91)
- GUI-Actor: Coordinate-Free Visual Grounding for GUI Agents (NeurIPS 2025 · similarity 0.90)
- FocusUI: Efficient UI Grounding via Position-Preserving Visual Token Selection (CVPR 2026 · similarity 0.89)
- DRS-GUI: Dynamic Region Search for Training-Free GUI Grounding (CVPR 2026 · similarity 0.89)
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.
