Research map / Vision-language and multimodal / Human and embodied modeling
Grounded Image Text Matching with Mismatched Relation Reasoning
Yu Wu, Yana Wei, Haozhe Wang, Yongfei Liu, Sibei Yang, Xuming He
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 |
|---|---|---|---|
| replace pretrained VL models with structured graph reasoning on language scene graph (RCRN) | UNITER-based matching · GITM-MR OOD (Train-Len11) | Match accuracy: 54.97 -> 63.19 | Improved |
| add bi-directional message propagation (ablation) | RCRN · GITM-MR Train-Len16 | Match accuracy: 60.07 -> 61.47 | Improved |
| add gating inputs from language and local similarity (Lang-beta, Ctx-beta) (ablation) | RCRN · GITM-MR Train-Len16 | Match accuracy: 59.65 -> 61.47 | Improved |
| add pretrained VL transformer layers (VLP) (ablation) | RCRN · GITM-MR Train-Len16 | MRR accuracy: 36.65 -> 53.23 | Improved |
| add pretrained VL transformer layers (VLP) (ablation) | RCRN · GITM-MR Train-Len16 | Grounding accuracy: 29.25 -> 23.37 | Hurt |
| add message passing over VLP-only model (ablation) | UNITER 6-layer with heads · GITM-MR Train-Len16 | Match accuracy: 54.79 -> 59.65 | Improved |
| pretrained models trained on limited data | TCL · GITM-MR in-distribution | Match accuracy: 53.11 (Len16) | No clear effect |
| test on out-of-distribution longer sentences | pretrained VL models (UNITER, FIBER) · GITM-MR Train-Len16 | Grounding accuracy: UNITER 40.63 -> 20.67; FIBER 37.15 -> 3.03 | Hurt |
Closest papers by meaning
- Visual Semantic Reasoning for Image-Text Matching (ICCV 2019 · similarity 0.88)
- Graph Structured Network for Image-Text Matching (CVPR 2020 · similarity 0.88)
- Composing Object Relations and Attributes for Image-Text Matching (CVPR 2024 · similarity 0.87)
- Relationship-Embedded Representation Learning for Grounding Referring Expressions (TPAMI 2020 · similarity 0.87)
- Learning Semantic Relationship Among Instances for Image-Text Matching (CVPR 2023 · similarity 0.87)
- Expanding the Scope of Negatives: Boosting Image-Text Matching with Negatives Distribution Guided Learning (AAAI 2025 · similarity 0.85)
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.
