Research map / Detection and segmentation / Open-set and few-shot detection
Deep Compositional Captioning: Describing Novel Object Categories Without Paired Training Data
Check 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 |
|---|---|---|---|
| add direct transfer from semantically similar known words (ablation) | DCC image captioning · MSCOCO held-out objects | F1: 0 -> 39.78 | Helped |
| add direct transfer from semantically similar known words (ablation) | DCC image captioning · MSCOCO held-out objects | BLEU-1: 62.99 -> 64.40 | Helped |
| add direct transfer from semantically similar known words (ablation) | DCC image captioning · MSCOCO held-out objects | METEOR: 19.9 -> 21.00 | Helped |
| use direct transfer rather than delta transfer (ablation) | DCC image captioning · MSCOCO held-out objects | F1: 34.89 -> 39.78 | Helped |
| use ImageNet images instead of MSCOCO images to train the lexical classifier (ablation) | DCC image captioning with MSCOCO language training · MSCOCO held-out objects | F1: 39.78 -> 33.60 | Hurt |
| use CaptionTxt instead of MSCOCO captions to train the language model, with ImageNet image training (ablation) | DCC image captioning · MSCOCO held-out objects | F1: 33.60 -> 35.53 | Helped |
| use WebCorpus instead of MSCOCO captions to train the language model, with ImageNet image training (ablation) | DCC image captioning · MSCOCO held-out objects | F1: 33.60 -> 34.94 | Helped |
| average the five closest weight vectors before delta transfer instead of using one (ablation) | DCC delta transfer · MSCOCO held-out objects | F1: 34.89 -> 34.60 | No clear effect |
| average the fifteen closest weight vectors before delta transfer instead of using one (ablation) | DCC delta transfer · MSCOCO held-out objects | F1: 34.89 -> 34.98 | No clear effect |
| add unpaired text pretraining for the language model (ablation) | DCC image captioning · MSCOCO held-out objects | sentence fluency | Helped |
| add direct transfer (ablation) | DCC video captioning · MSVD held-out objects | F1: 0.0 -> 6.0 | Helped |
| add ImageNet videos to lexical-classifier training, together with direct transfer (ablation) | DCC video captioning · MSVD held-out objects | F1: 6.0 -> 22.2 | Helped |
Closest papers by meaning
- Captioning Images With Diverse Objects (CVPR 2017 · similarity 0.93)
- Beyond a Pre-Trained Object Detector: Cross-Modal Textual and Visual Context for Image Captioning (CVPR 2022 · similarity 0.90)
- Incorporating Copying Mechanism in Image Captioning for Learning Novel Objects (CVPR 2017 · similarity 0.90)
- BACON: Improving Clarity of Image Captions via Bag-of-Concept Graphs (CVPR 2025 · similarity 0.87)
- Relational Distant Supervision for Image Captioning without Image-Text Pairs (AAAI 2024 · similarity 0.87)
- A Picture is Worth More Than 77 Text Tokens: Evaluating CLIP-Style Models on Dense Captions (CVPR 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.