Research map / Vision-language and multimodal / Human and embodied modeling
Paraphrasing Is All You Need for Novel Object Captioning
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 P2C instead of VinVL+VIVO | Novel object captioning · nocaps validation overall | CIDEr: 91.4 -> 97.2 | Helped |
| Add P2C instead of VinVL+VIVO | Novel object captioning · nocaps test overall | CIDEr: 88.3 -> 93.5 | Helped |
| Add language-model paraphrasing loss L_P (ablation) | P2C captioner · nocaps validation overall | CIDEr: 81.17 -> 84.20 | Helped |
| Add semantic-preserving gate g (ablation) | P2C captioner · nocaps validation overall | CIDEr: 21.3 -> 96.3 | Helped |
| Add image-text association reward r_A (ablation) | P2C captioner · nocaps validation overall | SPICE: 13.1 -> 14.1 | Helped |
| Add repetition penalty r_rep (ablation) | P2C captioner · nocaps validation overall | CIDEr: 89.1 -> 96.3 | Helped |
| Add repetition penalty r_rep (ablation) | P2C captioner · nocaps validation overall | SPICE: 13.9 -> 14.1 | No clear effect |
| Use BERT-large instead of BERT-base as the paraphrase model (ablation) | P2C Stage 1 · nocaps validation overall | CIDEr: 84.1 -> 84.2 | No clear effect |
| Replace CLIP with VIFIDEL as association model (ablation) | P2C Stage 2 · nocaps validation overall | CIDEr: 96.3 -> 55.3 | Hurt |
| Replace CLIP with SR-PL as association model (ablation) | P2C Stage 2 · nocaps validation overall | CIDEr: 96.3 -> 84.6 | Hurt |
| Use P2C instead of VinVL+VIVO | Novel object captioning · nocaps out-of-domain | object precision: 49.0 -> 51.3 | Mixed |
Closest papers by meaning
- BACON: Improving Clarity of Image Captions via Bag-of-Concept Graphs (CVPR 2025 · similarity 0.87)
- CAPability: A Comprehensive Visual Caption Benchmark for Evaluating Both Correctness and Thoroughness (NeurIPS 2025 · similarity 0.87)
- Panoptic Captioning: An Equivalence Bridge for Image and Text (NeurIPS 2025 · similarity 0.86)
- Mining Fine-Grained Image-Text Alignment for Zero-Shot Captioning via Text-Only Training (AAAI 2024 · similarity 0.86)
- LLMDet: Learning Strong Open-Vocabulary Object Detectors under the Supervision of Large Language Models (CVPR 2025 · similarity 0.86)
- EVCap: Retrieval-Augmented Image Captioning with External Visual-Name Memory for Open-World Comprehension (CVPR 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.