Research map / Detection and segmentation / Open-set and few-shot detection
Generation and Comprehension of Unambiguous Object Descriptions
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 maximum-likelihood training with max-margin MMI training using hard multibox negatives | VGGNet + LSTM referring-expression model · G-Ref-Test, multibox proposals, GT descriptions | precision@1: 0.406 -> 0.446 | Helped |
| Replace maximum-likelihood training with max-margin MMI training using hard multibox negatives | VGGNet + LSTM referring-expression model · UNC-Ref-Test, multibox proposals, GT descriptions | precision@1: 0.477 -> 0.518 | Helped |
| Use max-margin MMI training rather than maximum-likelihood training | VGGNet + LSTM referring-expression model · G-Ref-Test, GT proposals, GT descriptions | precision@1: 0.545 -> 0.606 | Helped |
| Use max-margin MMI training rather than maximum-likelihood training | VGGNet + LSTM referring-expression model · UNC-Ref-Test, GT proposals, GT descriptions | precision@1: 0.643 -> 0.700 | Helped |
| Train with hard rather than easy ground-truth negatives (ablation) | MMI max-margin model · UNC-Ref-Val, GT proposals, generated descriptions | precision@1: 0.851 -> 0.857 | Helped |
| Use hard multibox negatives instead of hard ground-truth negatives (ablation) | MMI max-margin model · UNC-Ref-Val, multibox proposals, generated descriptions | precision@1: 0.591 -> 0.604 | Helped |
| Replace max-margin MMI with softmax MMI (ablation) | MMI model with hard multibox negatives · UNC-Ref-Val, GT proposals, generated descriptions | precision@1: 0.848 -> 0.848 | No clear effect |
| Add automatically described bounding-box-only data to strongly labeled training data (ablation) | Full Model · G-Ref, GT proposals, generated descriptions | precision@1: 0.791 -> 0.793 | Helped |
| Add automatically described bounding-box-only data to strongly labeled training data (ablation) | Full Model · UNC-Ref, multibox proposals, GT descriptions | precision@1: 0.483 -> 0.486 | Helped |
| Evaluate generated descriptions rather than human descriptions | Full Model comprehension system · UNC-Ref-Val, GT proposals | precision@1: 0.695 vs 0.848 | Helped |
| Use the full model instead of the baseline for generated descriptions | G-Ref-Test human evaluation · G-Ref-Test | better or equal to human caption: 15.9% -> 20.4% | Helped |
Closest papers by meaning
- Rex-Thinker: Grounded Object Referring via Chain-of-Thought Reasoning (ICLR 2026 · similarity 0.88)
- RefDetector: A Simple Yet Effective Matching-based Method for Referring Expression Comprehension (AAAI 2025 · similarity 0.87)
- Referring Image Editing: Object-level Image Editing via Referring Expressions (CVPR 2024 · similarity 0.87)
- Unleashing Region Understanding in Intermediate Layers for MLLM-based Referring Expression Generation (NeurIPS 2024 · similarity 0.86)
- Unveiling Parts Beyond Objects: Towards Finer-Granularity Referring Expression Segmentation (CVPR 2024 · similarity 0.86)
- VisMin: Visual Minimal-Change Understanding (NeurIPS 2024 · 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.