Research map / Detection and segmentation / Open-set and few-shot detection
LoA-Trans: Enhancing Visual Grounding by Location-Aware Transformers
Ziling Huang, Shin'ichi Satoh
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 TaskSyn Network for task information exchange in decoder (ablation) | LoA-Trans-S · RefCOCO val | mIoU: 74.58 -> 76.03 | Improved |
| add TaskSyn Network (ablation) | LoA-Trans-S · RefCOCO val | Pr@0.5: 86.52 -> 87.59 | Improved |
| use query selection instead of random initialized queries (ablation) | LoA-Trans-S · RefCOCO val | mIoU: 75.67 -> 76.03 | Improved |
| use query selection instead of using all features as queries (ablation) | LoA-Trans-S · RefCOCO val | memory: Out of Memory -> 87.59 Pr@0.5 | Improved |
| increase decoder layers from 3 to 6 (ablation) | LoA-Trans-S · RefCOCO val | Pr@0.5: 87.20 -> 87.59 | Improved |
| increase decoder layers from 6 to 8 (ablation) | LoA-Trans-S · RefCOCO val | Pr@0.5: 87.59 -> 86.76 | Hurt |
| increase selected queries K from 50 to 100 (ablation) | LoA-Trans-S · RefCOCO val | Pr@0.5: 86.80 -> 87.59 | Improved |
| increase selected queries K from 100 to 125 (ablation) | LoA-Trans-S · RefCOCO val | Pr@0.5: 87.59 -> 86.77 | Hurt |
| replace estimated center with ground-truth center (ablation) | LoA-Trans-B · RefCOCO val, centerness [0.0,0.1) | Pr@0.5: 10.95 -> 34.40 | Improved |
| use LoA-Trans-B instead of VG-LAW on RefCOCO+ testB segmentation | Visual grounding · RefCOCO+ testB (RES) | mIoU: 58.89 -> 62.90 | Improved |
| use LoA-Trans-S instead of VG-LAW on RefCOCO testB (REC) | Visual grounding · RefCOCO testB (REC) | Pr@0.5: 83.16 (VG-LAW) vs 82.98 (LoA-Trans-S) | No clear effect |
Closest papers by meaning
- Referencing Where to Focus: Improving Visual Grounding with Referential Query (NeurIPS 2024 · similarity 0.91)
- Your Large Vision-Language Model Only Needs A Few Attention Heads For Visual Grounding (CVPR 2025 · similarity 0.90)
- Multi-task Visual Grounding with Coarse-to-Fine Consistency Constraints (AAAI 2025 · similarity 0.89)
- Mask Grounding for Referring Image Segmentation (CVPR 2024 · similarity 0.89)
- UGround: Towards Unified Visual Grounding with Unrolled Transformers (ICML 2026 · similarity 0.88)
- Knowing Your Target: Target-Aware Transformer Makes Better Spatio-Temporal Video Grounding (ICLR 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.
