Research map / Vision-language and multimodal / CLIP and zero-shot
Role-SynthCLIP: A Role-Play Driven Diverse Synthetic Data Approach
Yuanxiang Huangfu, Chaochao Wang, Weilei Wang
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 the interpreter expert role (ablation) | Role-SynthCLIP (CLIP-L/14) · MS-COCO image-to-text | Recall@1: 67.0 -> 68.6 | Improved |
| add the emotion expert role (ablation) | Role-SynthCLIP (CLIP-L/14) · MS-COCO image-to-text | Recall@1: 67.7 -> 68.6 | Improved |
| add the compositional analyst expert role (ablation) | Role-SynthCLIP (CLIP-L/14) · MS-COCO image-to-text | Recall@1: 67.4 -> 68.6 | Improved |
| add the narrative setter expert role (ablation) | Role-SynthCLIP (CLIP-L/14) · MS-COCO image-to-text | Recall@1: 67.6 -> 68.6 | Improved |
| add the observer expert role (ablation) | Role-SynthCLIP (CLIP-L/14) · MS-COCO image-to-text | Recall@1: 66.7 -> 68.6 | Improved |
| generate short captions directly with role prompts instead of first sentence of long caption (ablation) | Role-SynthCLIP · MS-COCO image-to-text | Recall@1: 58.7 -> 68.6 | Improved |
| generate short captions directly instead of LLM summarization (ablation) | Role-SynthCLIP · MS-COCO image-to-text | Recall@1: 60.1 -> 68.6 | Improved |
| generate short captions directly instead of random sentence extraction (ablation) | Role-SynthCLIP · MS-COCO image-to-text | Recall@1: 56.1 -> 68.6 | Improved |
| use role-play prompts instead of multiple generic prompts (ablation) | Role-SynthCLIP · MS-COCO image-to-text | Recall@1: 64.3 -> 68.6 | Improved |
| add the Role-Aware Filter instead of random subsampling (ablation) | Role-SynthCLIP · MS-COCO image-to-text | Recall@1: 67.0 -> 68.6 | Improved |
| add the Role-Aware Filter instead of random subsampling (ablation) | Role-SynthCLIP · Flickr-1k image-to-text | Recall@1: 90.1 -> 92.1 | Improved |
| use Role-SynthCLIP 1M data instead of Long-CLIP 1M data | CLIP-B/16 · Retrieval average | Recall@1 average: 69.53 -> 77.01 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- Long-CLIP 1M — Long-CLIP: Unlocking the Long-Text Capability of CLIP (ECCV 2024) · better on 3 of 4 shared results
- FIX-CLIP 1M — Fix-CLIP: Dual-Branch Hierarchical Contrastive Learning via Synthetic Captions for Better Understanding of Long Text (ICCV 2025) · better on 2 of 2 shared results
- LoTLIP 100M — LoTLIP: Improving Language-Image Pre-training for Long Text Understanding (NeurIPS 2024) · 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
- Enhancing Vision-Language Compositional Understanding with Multimodal Synthetic Data (CVPR 2025 · similarity 0.88)
- Image Captioning with Multi-Context Synthetic Data (AAAI 2024 · similarity 0.87)
- CapsFusion: Rethinking Image-Text Data at Scale (CVPR 2024 · similarity 0.87)
- CtrlSynth: Controllable Image Text Synthesis for Data-Efficient Multimodal Learning (ICML 2025 · similarity 0.87)
- SmartCLIP: Modular Vision-language Alignment with Identification Guarantees (CVPR 2025 · similarity 0.86)
- Revisit Large-Scale Image-Caption Data in Pre-training Multimodal Foundation Models (ICLR 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.
