Research map / Vision-language and multimodal / Human and embodied modeling
Relationship-Embedded Representation Learning for Grounding Referring Expressions
Sibei Yang, Guanbin Li, Yizhou Yu
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 language-guided visual context (guided viscxt) instead of global visual context (ablation) | CMRIN · RefCOCOg test | Precision@1: 75.23 -> 78.29 | Improved |
| add global graph visual context over instance-only baseline (ablation) | CMRIN · RefCOCOg test | Precision@1: 71.26 -> 75.23 | Improved |
| fuse word context into the language-guided graph (multimodal fusion) (ablation) | CMRIN · RefCOCOg test | Precision@1: 78.29 -> 80.85 | Improved |
| use 2 gated graph conv layers instead of 1 (ablation) | CMRIN · RefCOCOg test | Precision@1: 79.38 -> 80.85 | Improved |
| use 3 gated graph conv layers instead of 2 (ablation) | CMRIN · RefCOCOg test | Precision@1: 80.85 -> 80.55 | No clear effect |
| train with triplet loss with hard negative mining instead of softmax loss (ablation) | CMRIN · RefCOCOg test | Precision@1: 79.37 -> 80.85 | Improved |
| use margin 0.5 instead of 0.1 in triplet loss (ablation) | CMRIN · RefCOCO+ val | Precision@1: 75.63 -> 74.52 | Hurt |
| vary negative sampling strategy (hardest, semi-hard, two negatives) (ablation) | CMRIN · RefCOCO val | Precision@1: 86.43 -> 86.38 | No clear effect |
| add explicit noun phrases for vertex highlighting (ablation) | CMRIN · RefCOCO+ testB | Precision@1: 68.23 -> 69.58 | Improved |
| use 11 edge types instead of 7 types (ablation) | CMRIN · RefCOCO testB | Precision@1: 84.26 -> 85.52 | Improved |
| connect pairs by center distance 0.7 instead of 0.5 (ablation) | CMRIN · RefCOCOg test | Precision@1: 80.85 -> 80.20 | Hurt |
| add semantic relation graph branch to spatial branch (ablation) | CMRIN · RefCOCO testA | Precision@1: 87.36 -> 88.17 | 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 9 of 9 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
- Cross-Modal Relationship Inference for Grounding Referring Expressions (CVPR 2019 · similarity 0.97)
- Graph-Structured Referring Expression Reasoning in the Wild (CVPR 2020 · similarity 0.91)
- Improving Referring Expression Grounding With Cross-Modal Attention-Guided Erasing (CVPR 2019 · similarity 0.91)
- Grounding Referring Expressions in Images by Variational Context (CVPR 2018 · similarity 0.91)
- RefDetector: A Simple Yet Effective Matching-based Method for Referring Expression Comprehension (AAAI 2025 · similarity 0.90)
- Referencing Where to Focus: Improving Visual Grounding with Referential Query (NeurIPS 2024 · 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.
