Research map / Detection and segmentation / Open-set and few-shot detection
NOC-REK: Novel Object Captioning With Retrieved Vocabulary From External Knowledge
Duc Minh Vo, Hong Chen, Akihiro Sugimoto, Hideki Nakayama
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 |
|---|---|---|---|
| Use NOC-REK instead of the strongest reported baseline | Novel object captioning · Held-out COCO novel classes | Avg. F1-score: 57.9 -> 76.3 | Improved |
| Use NOC-REK instead of the strongest reported baseline | Image captioning · Nocaps validation Overall | CIDEr: 94.3 -> 100.9 | Improved |
| Use NOC-REK instead of the strongest reported baseline | Image captioning · Nocaps test Overall | CIDEr: 92.5 -> 93.0 | Improved |
| Add second-stage SCST/CIDEr training and the reward for retrieved vocabulary appearing in captions (ablation) | NOC-REK* · Held-out COCO novel classes | Avg. F1-score: 67.7 -> 76.3 | Improved |
| Add second-stage SCST/CIDEr training and the reward for retrieved vocabulary appearing in captions (ablation) | NOC-REK* · Held-out COCO | CIDEr: 109.3 -> 138.4 | Improved |
| Add second-stage SCST/CIDEr training and the reward for retrieved vocabulary appearing in captions (ablation) | NOC-REK* · Nocaps validation Overall | CIDEr: 82.9 -> 100.9 | Improved |
| Add second-stage SCST/CIDEr training and the reward for retrieved vocabulary appearing in captions (ablation) | NOC-REK* · Nocaps validation Overall | SPICE: 12.0 -> 14.0 | Improved |
| Use the full COCO + Nocaps external vocabulary instead of removing 25% at random (ablation) | NOC-REK · Nocaps validation Overall | CIDEr: 85.8 -> 100.9 | Improved |
| Use the full COCO + Nocaps external vocabulary instead of removing 50% at random (ablation) | NOC-REK · Nocaps validation Overall | CIDEr: 83.0 -> 100.9 | Improved |
| Use the full COCO + Nocaps external vocabulary instead of removing 75% at random (ablation) | NOC-REK · Nocaps validation Overall | CIDEr: 83.2 -> 100.9 | Improved |
| Add COCO vocabulary to external knowledge (ablation) | NOC-REK without COCO vocabulary · Nocaps validation Overall | CIDEr: 82.9 -> 100.9 | Improved |
| Add Nocaps vocabulary to external knowledge (ablation) | NOC-REK without Nocaps vocabulary · Nocaps validation Overall | CIDEr: 82.4 -> 100.9 | Improved |
Closest papers by meaning
- Captioning Images With Diverse Objects (CVPR 2017 · similarity 0.90)
- EVCap: Retrieval-Augmented Image Captioning with External Visual-Name Memory for Open-World Comprehension (CVPR 2024 · similarity 0.87)
- Cyclic Contrastive Knowledge Transfer for Open-Vocabulary Object Detection (ICLR 2025 · similarity 0.86)
- LLMDet: Learning Strong Open-Vocabulary Object Detectors under the Supervision of Large Language Models (CVPR 2025 · similarity 0.86)
- Relational Distant Supervision for Image Captioning without Image-Text Pairs (AAAI 2024 · similarity 0.85)
- DetCLIPv3: Towards Versatile Generative Open-vocabulary Object Detection (CVPR 2024 · 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.
