Research map / Vision-language and multimodal / Human and embodied modeling
Small Object, Great Challenge: A Benchmark for Small Object Visual Grounding
Wenqi Jia, Ruifan Li, Pengyue Lin, Fangxiang Feng, Zhanyu Ma, Xiaojie Wang
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 Hierarchical Textual Infusion (HTI) module to the visual encoder (ablation) | SoVG-Net baseline · RefCOCOs val (REC) | Acc@0.5: 68.61 -> 72.69 | Improved |
| add HTI module (ablation) | SoVG-Net baseline · RefCOCOs testB (REC) | Acc@0.5: 66.82 -> 72.33 | Improved |
| add HTI module (ablation) | SoVG-Net baseline · RefCOCOs val (RES) | mIoU: 56.10 -> 60.58 | Improved |
| move text infusion from shallow blocks to deep blocks (15,18,21,24) (ablation) | HTI module · RefCOCOs testA (REC) | Acc@0.5: 51.83 -> 63.39 | Improved |
| use uniform infusion instead of middle infusion (ablation) | HTI module · RefCOCOs val (REC) | Acc@0.5: 71.49 -> 71.23 | No clear effect |
| train on RefCOCOs instead of using models without RefCOCOs training (ablation) | TransVG · RefCOCOs testB (REC) | Acc@0.5: 11.71 -> 64.23 | Improved |
| train on RefCOCOs instead of RefCOCOg (qualitative) (ablation) | SoVG-Net · small object grounding examples | qualitative localization | Improved |
| use SoVG-Net instead of EEVG on RefCOCOs testB | small object grounding · RefCOCOs testB (REC) | Acc@0.5: 66.76 -> 72.33 | Improved |
| use SoVG-Net on classic RefCOCOg instead of EEVG | classic VG · RefCOCOg test (U) (REC) | Accuracy: 80.24 -> 79.13 | Hurt |
| use SoVG-Net on classic RefCOCO val instead of EEVG | classic VG · RefCOCO val (REC) | Accuracy: 88.08 -> 89.46 | Improved |
Closest papers by meaning
- When Visual Grounding Meets Gigapixel-level Large-scale Scenes: Benchmark and Approach (CVPR 2024 · similarity 0.90)
- GroundVLP: Harnessing Zero-Shot Visual Grounding from Vision-Language Pre-training and Open-Vocabulary Object Detection (AAAI 2024 · similarity 0.90)
- SeeGround: See and Ground for Zero-Shot Open-Vocabulary 3D Visual Grounding (CVPR 2025 · similarity 0.89)
- From Objects to Anywhere: A Holistic Benchmark for Multi-level Visual Grounding in 3D Scenes (NeurIPS 2025 · similarity 0.89)
- Visual Programming for Zero-shot Open-Vocabulary 3D Visual Grounding (CVPR 2024 · similarity 0.88)
- Referencing Where to Focus: Improving Visual Grounding with Referential Query (NeurIPS 2024 · 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.
