Research map / Vision-language and multimodal / CLIP and zero-shot
Zero-shot Visual Relation Detection via Composite Visual Cues from Large Language Models
Read the paperCheck 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 |
|---|---|---|---|
| Use composite subject, object, and spatial descriptions with dynamic cue weights | CLIP for zero-shot VRD · VG | R@100: 13.2 -> 19.3 | Helped |
| Use composite subject, object, and spatial descriptions with dynamic cue weights | CLIP for zero-shot VRD · GQA | R@100: 9.9 -> 11.8 | Helped |
| Add high-level object-category guidance when generating component cues (ablation) | RECODE · VG | R@100: 15.4 -> 19.3 | Helped |
| Add high-level object-category guidance when generating component cues (ablation) | RECODE · GQA | R@100: 10.2 -> 11.8 | Helped |
| Use relation-class descriptions instead of the CLS prompt (ablation) | CLIP zero-shot VRD · VG | R@100: 13.2 -> 12.9 | Hurt |
| Use relation-class descriptions instead of the CLS prompt (ablation) | CLIP zero-shot VRD · GQA | R@100: 9.9 -> 9.3 | Hurt |
| Add GPT filtering of unreasonable predictions (ablation) | RECODE · VG | R@100: 19.3 -> 25.0 | Helped |
| Add GPT filtering of unreasonable predictions (ablation) | RECODE · GQA | R@100: 11.8 -> 15.4 | Helped |
| Add description-based cues to the CLS baseline (ablation) | RECODE · VG | R@100: 13.2 -> 17.4 | Helped |
| Add spatial features to cue-based RECODE (ablation) | RECODE · VG | R@100: 17.4 -> 19.3 | Helped |
| Use dynamic LLM cue weights instead of average aggregation (ablation) | RECODE · VG | R@100: +0.5% to 1.1% gains on R@K | Helped |
| Use RECODE rather than CLS | Zero-shot VRD · HICO-DET | mAP (reported column 1): 32.3 -> 32.7 | Helped |
Closest papers by meaning
- Multi-Modal Prompting for Open-Vocabulary Video Visual Relationship Detection (AAAI 2024 · similarity 0.89)
- Visual Diversity and Region-aware Prompt Learning for Zero-shot HOI Detection (NeurIPS 2025 · similarity 0.89)
- Zero-shot Referring Expression Comprehension via Structural Similarity Between Images and Captions (CVPR 2024 · similarity 0.88)
- TPR: Topology-Preserving Reservoirs for Generalized Zero-Shot Learning (NeurIPS 2024 · similarity 0.88)
- Interpreting and Analysing CLIP's Zero-Shot Image Classification via Mutual Knowledge (NeurIPS 2024 · similarity 0.87)
- GroundVLP: Harnessing Zero-Shot Visual Grounding from Vision-Language Pre-training and Open-Vocabulary Object Detection (AAAI 2024 · similarity 0.87)
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.