Research map / Vision-language and multimodal / CLIP and zero-shot
RCA-NOC: Relative Contrastive Alignment for Novel Object Captioning
Jiashuo Fan, Yaoyuan Liang, Leyao Liu, Shaolun Huang, Lei Zhang
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 cross-modality alignment contrastive loss (ablation) | RCA-NOC · NoCaps val | CIDEr: 90.9 -> 96.5 | Improved |
| Add inner-modality alignment (ablation) | RCA-NOC · NoCaps val | CIDEr: 96.5 -> 98.7 | Improved |
| Add uncertainty-aware selection and re-weighting (UASR) (ablation) | RCA-NOC · NoCaps val | CIDEr: 98.7 -> 103.2 | Improved |
| Extract augmented tags with CLIP (ablation) | RCA-NOC · NoCaps val | CIDEr: 103.2 -> 107.1 | Improved |
| Generate augmented tags with NOC-REK instead of CLIP (ablation) | RCA-NOC · NoCaps val | CIDEr: 103.2 -> 103.4 | No clear effect |
| Use CLIP tags instead of VIVO tags (ablation) | RCA-NOC · NoCaps val | CIDEr: 105.2 -> 107.1 | Improved |
| Add Uncertainty-Aware Re-weighting only (ablation) | RCA-NOC · NoCaps val | CIDEr: 102.4 -> 105.6 | Improved |
| Filter false negatives with Uncertainty-Aware Selection for Negatives (ablation) | RCA-NOC · NoCaps val | CIDEr: 102.4 -> 104.2 | Improved |
| Apply RCA-NOC to general captioning pretraining | RCA-NOC · COCO Karpathy test | CIDEr: 118.3 -> 128.4 | Improved |
| Replace NOC-REK with RCA-NOC on Held-Out COCO | RCA-NOC · Held-Out COCO test | CIDEr: 138.4 -> 142.8 | Improved |
Closest papers by meaning
- SyCoCa: Symmetrizing Contrastive Captioners with Attentive Masking for Multimodal Alignment (ICML 2024 · similarity 0.87)
- SmartCLIP: Modular Vision-language Alignment with Identification Guarantees (CVPR 2025 · similarity 0.87)
- Mining Fine-Grained Image-Text Alignment for Zero-Shot Captioning via Text-Only Training (AAAI 2024 · similarity 0.86)
- Seeing the Image: Prioritizing Visual Correlation by Contrastive Alignment (NeurIPS 2024 · similarity 0.86)
- SuperCLIP: CLIP with Simple Classification Supervision (NeurIPS 2025 · similarity 0.85)
- GOAL: Global-local Object Alignment Learning (CVPR 2025 · similarity 0.85)
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.
