Research map / Vision-language and multimodal / Human and embodied modeling
Understanding Synonymous Referring Expressions via Contrastive Features
Yiwen Chen, Yi–Hsuan Tsai, Ming–Hsuan Yang
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 synonymous-sentence contrastive feature losses | Ours (LSTM) · RefCOCO val | accuracy: 76.63 -> 79.09 | Improved |
| add synonymous-sentence contrastive feature losses | Ours (BERT) · RefCOCO val | accuracy: 77.02 -> 79.63 | Improved |
| add losses when jointly training on three datasets | Ours (BERT) · RefCOCO val joint | accuracy: 75.17 -> 82.42 | Improved |
| add image-level feature learning only (ablation) | baseline · RefCOCO val joint | accuracy: 75.17 -> 77.82 | Improved |
| add instance-level contrastive loss only (ablation) | baseline · RefCOCO val joint | accuracy: 75.17 -> 79.33 | Improved |
| use contrastive instead of triplet loss at instance level (ablation) | final model · RefCOCO val joint | accuracy: 80.61 -> 82.42 | Improved |
| add negative mining (ablation) | final model · RefCOCO val joint | accuracy: 82.05 -> 82.42 | Improved |
| add GCN (ablation) | final model · RefCOCO val joint | accuracy: 81.40 -> 82.42 | Improved |
| train with losses on RefCOCOg, evaluate unseen on Ref-Reasoning one-object | Ours · Ref-Reasoning (one object) | accuracy: 50.26 -> 53.49 | Improved |
| plain pre-training on RefCOCO datasets without proposed loss (ablation) | Ours (baseline) · Ref-Reasoning one-object | accuracy: 76.43 -> 76.54 | No clear effect |
| add loss in both pre-training and fine-tuning (ablation) | Ours · Ref-Reasoning one-object | accuracy: 76.54 -> 81.94 | Improved |
| add loss in both pre-training and fine-tuning (ablation) | Ours · ReferItGame test | accuracy: 57.13 -> 61.71 | 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 1 of 1 shared result
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
- RefDetector: A Simple Yet Effective Matching-based Method for Referring Expression Comprehension (AAAI 2025 · similarity 0.90)
- Revisiting Counterfactual Problems in Referring Expression Comprehension (CVPR 2024 · similarity 0.90)
- Rethinking Two-Stage Referring Expression Comprehension: A Novel Grounding and Segmentation Method Modulated by Point (AAAI 2024 · similarity 0.89)
- ScanFormer: Referring Expression Comprehension by Iteratively Scanning (CVPR 2024 · similarity 0.89)
- Ref-Adv: Exploring MLLM Visual Reasoning in Referring Expression Tasks (ICLR 2026 · similarity 0.89)
- Zero-shot Referring Expression Comprehension via Structural Similarity Between Images and Captions (CVPR 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.
