Research map / Vision-language and multimodal / Human and embodied modeling
Improving One-stage Visual Grounding by Recursive Sub-query Construction
Zhengyuan Yang, Tianlang Chen, Liwei Wang, Jiebo Luo
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 recursive sub-query learner referring to visual features (ablation) | Average vector query · RefCOCOg val-g (ablation) | Acc@0.5: 59.24 -> 60.96 | Improved |
| add visual-conditioned sub-query learner over text-only attention (ablation) | Single-head attention · RefCOCOg val-g (ablation) | Acc@0.5: 59.43 -> 60.96 | Improved |
| use text-only multi-head attention (ablation) | Average vector query · RefCOCOg val-g (ablation) | Acc@0.5: 59.24 -> 59.25 | No clear effect |
| use per-word sub-queries (ablation) | Average vector query · RefCOCOg val-g (ablation) | Acc@0.5: 59.24 -> 58.36 | Hurt |
| build sub-queries independently at each spatial location (ablation) | Sub-query learner · RefCOCOg val-g (ablation) | Acc@0.5: 60.96 -> 60.81 | No clear effect |
| use sub-query modulation instead of Concat-Conv fusion (ablation) | Concat-Conv fusion · RefCOCOg val-g (ablation) | Acc@0.5: 59.20 -> 60.96 | Improved |
| increase number of rounds K from 1 to 3 (ablation) | Ours-Base · RefCOCOg val-g (ablation) | Acc@0.5: 58.22 -> 60.96 | Improved |
| increase number of rounds K from 3 to 4 (ablation) | Ours-Base · RefCOCOg val-g (ablation) | Acc@0.5: 60.96 -> 61.08 | No clear effect |
| add ConvLSTM grounding module (ablation) | Ours-Base · RefCOCOg val-g (ablation) | Acc@0.5: 60.96 -> 61.26 | Improved |
| increase input size to 512 (ablation) | Ours-Base · RefCOCOg val-g (ablation) | Acc@0.5: 60.96 -> 61.99 | Improved |
| add both ConvLSTM and size 512 (ablation) | Ours-Base · RefCOCOg val-g (ablation) | Acc@0.5: 60.96 -> 63.12 | Improved |
| use Ours-Base instead of One-Stage-BERT on longest RefCOCO queries (6+ words) | One-Stage-BERT · RefCOCO (queries with 6+ words) | Acc@0.5: 55.59 -> 66.19 | 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.
- MAttNet — MAttNet: Modular Attention Network for Referring Expression Comprehension (CVPR 2018) · better on 2 of 2 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
- Referencing Where to Focus: Improving Visual Grounding with Referential Query (NeurIPS 2024 · similarity 0.91)
- SeqTR: A Simple Yet Universal Network for Visual Grounding (ECCV 2022 · similarity 0.89)
- Referring Transformer: A One-step Approach to Multi-task Visual Grounding (NeurIPS 2021 · similarity 0.89)
- Connecting the Dots: Training-Free Visual Grounding via Agentic Reasoning (AAAI 2026 · similarity 0.89)
- Enhancing Part-Level Point Grounding for Any Open-Source MLLMs (CVPR 2026 · similarity 0.87)
- GroundingME: Exposing the Visual Grounding Gap in MLLMs through Multi-Dimensional Evaluation (CVPR 2026 · similarity 0.87)
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.
