Research map / Detection and segmentation / Open-set and few-shot detection
Modeling Context Between Objects for Referring Expression Understanding
Varun K. Nagaraja, Vlad I. Morariu, Larry S. Davis
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 |
|---|---|---|---|
| train with MIL negative bag margin loss instead of max-margin | referring expression LSTM · Google RefExp Val (GT proposals) | Precision@1: 65.7 -> 68.4 | Improved |
| train with MIL negative bag margin loss instead of max-margin | referring expression LSTM · UNC RefExp Val (GT proposals) | Precision@1: 74.4 -> 76.9 | Improved |
| train with MIL negative bag margin loss instead of max-margin | referring expression LSTM · UNC RefExp TestB (MCG proposals) | Precision@1: 55.1 -> 56.4 | Improved |
| train with MIL negative bag margin loss instead of max-margin | referring expression LSTM · UNC RefExp TestA (MCG proposals) | Precision@1: 58.4 -> 58.6 | No clear effect |
| add positive bag margin to negative bag margin | MIL loss · Google RefExp Val (MCG proposals) | Precision@1: 49.5 -> 50.0 | Improved |
| add positive bag margin to negative bag margin | MIL loss · UNC RefExp Val (GT proposals) | Precision@1: 76.9 -> 76.1 | Hurt |
| pool context with noisy-or instead of max (ablation) | MIL negative bag margin · Google RefExp Val (GT proposals) | Precision@1: 66.5 -> 68.4 | Improved |
| pool context over all regions with noisy-or instead of image only (ablation) | MIL negative bag margin · Google RefExp Val (GT proposals) | Precision@1: 65.9 -> 68.4 | Improved |
| pool context over all regions with noisy-or instead of image only (ablation) | MIL negative bag margin · UNC RefExp Val (GT proposals) | Precision@1: 76.4 -> 76.9 | Improved |
| pool context over all regions with noisy-or instead of image only (ablation) | MIL negative bag margin · UNC RefExp TestA (GT proposals) | Precision@1: 76.2 -> 75.6 | Hurt |
| train with MIL loss using image-only context instead of max-margin | referring expression LSTM · Google RefExp Val (GT proposals) | Precision@1: 65.7 -> 65.9 | Improved |
| use ground-truth proposals instead of MCG proposals | Ours, Pos. & Neg. Bag Mgn. · Google RefExp Val | Precision@1: 50.0 -> 68.4 | Improved |
Closest papers by meaning
- Modeling Context in Referring Expressions (ECCV 2016 · similarity 0.91)
- Understanding Synonymous Referring Expressions via Contrastive Features (IJCV 2022 · similarity 0.89)
- Rethinking Two-Stage Referring Expression Comprehension: A Novel Grounding and Segmentation Method Modulated by Point (AAAI 2024 · similarity 0.86)
- Ref-Adv: Exploring MLLM Visual Reasoning in Referring Expression Tasks (ICLR 2026 · similarity 0.86)
- Unveiling Parts Beyond Objects: Towards Finer-Granularity Referring Expression Segmentation (CVPR 2024 · similarity 0.86)
- RefDetector: A Simple Yet Effective Matching-based Method for Referring Expression Comprehension (AAAI 2025 · similarity 0.86)
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.
