Research map / Vision-language and multimodal / Human and embodied modeling
Variational Context: Exploiting Visual and Textual Context for Grounding Referring Expressions
Yulei Niu, Hanwang Zhang, Zhiwu Lu, Shih‐Fu Chang
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 KL divergence context regularization (ablation) | VC · RefCOCO+ Test A (ground-truth regions, supervised) | grounding accuracy (%): 60.76 -> 62.56 | Improved |
| add KL divergence context regularization (ablation) | VC · RefCOCOg Val* (ground-truth regions, supervised) | grounding accuracy (%): 71.05 -> 73.98 | Improved |
| add cue-specific word attention (ablation) | VC · RefCOCOg Val* (ground-truth regions, supervised) | grounding accuracy (%): 65.13 -> 73.98 | Improved |
| add cue-specific word attention (ablation) | VC · RefCOCO+ Test B (detected regions, supervised) | grounding accuracy (%): 46.88 -> 53.18 | Improved |
| add KL divergence context regularization (ablation) | VC · RefCOCO+ Test B (detected regions, supervised) | grounding accuracy (%): 51.30 -> 53.18 | Improved |
| add KL divergence context regularization (ablation) | VC · RefCOCOg Val* (unsupervised, detected regions) | grounding accuracy (%): 28.14 -> 33.66 | Improved |
| add KL divergence context regularization (ablation) | VC · RefCOCO+ Test A (unsupervised, detected regions) | grounding accuracy (%): 19.74 -> 25.79 | Improved |
| add cue-specific word attention (ablation) | VC · RefCOCO Test A (unsupervised, ground-truth regions) | grounding accuracy (%): 33.29 -> 17.34 | Hurt |
| add cue-specific word attention (ablation) | VC · RefCOCOg Val* (unsupervised, ground-truth regions) | grounding accuracy (%): 30.26 -> 33.79 | Improved |
| add referring expression generation module with REINFORCE policy gradient (ablation) | VC · RefCOCOg Val* (ground-truth regions, supervised) | grounding accuracy (%): 73.98 -> 74.96 | Improved |
| add generation module without policy gradient (ablation) | VC · RefCOCO Test B (ground-truth regions, supervised) | grounding accuracy (%): 82.39 -> 82.04 | Hurt |
| use ResNet-101 visual features instead of VGG | VC w/ Gen+PG · RefCOCO+ Test A (ground-truth regions, supervised) | grounding accuracy (%): 63.22 -> 67.52 | 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 4 of 5 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
- Ref-Adv: Exploring MLLM Visual Reasoning in Referring Expression Tasks (ICLR 2026 · similarity 0.89)
- Referencing Where to Focus: Improving Visual Grounding with Referential Query (NeurIPS 2024 · similarity 0.89)
- RefDetector: A Simple Yet Effective Matching-based Method for Referring Expression Comprehension (AAAI 2025 · similarity 0.89)
- Zero-shot Referring Expression Comprehension via Structural Similarity Between Images and Captions (CVPR 2024 · similarity 0.88)
- Unveiling Parts Beyond Objects: Towards Finer-Granularity Referring Expression Segmentation (CVPR 2024 · similarity 0.88)
- Boosting Weakly Supervised Referring Image Segmentation via Progressive Comprehension (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.
