Research map / Detection and segmentation / Open-set and few-shot detection
Captioning Images With Diverse Objects
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 |
|---|---|---|---|
| Use NOC instead of DCC | Image captioning of held-out MSCOCO objects · MSCOCO held-out objects (8-object split) | Avg. F1: 39.78 -> 48.79 | Helped |
| Use NOC instead of DCC | Image captioning of ImageNet novel objects · ImageNet novel objects (638 categories) | F1 (%): 14.47 -> 33.76 | Helped |
| Add external image and text training to a COCO-only caption model | Captioning rare ImageNet objects · ImageNet rare objects (52 categories) | Rare-word incorporation (%): 2.93% -> 35.15% | Helped |
| Add GloVe embeddings and LM pre-training (ablation) | NOC · MSCOCO held-out objects (8-object split) | F1: 14.41 -> 25.38 | Helped |
| Add the auxiliary objective (ablation) | NOC with LM and GloVe · MSCOCO held-out objects (8-object split) | F1: 25.38 -> 48.79 | Helped |
| Add LM pre-training (ablation) | NOC with GloVe, tuned vision, and auxiliary objective · MSCOCO held-out objects (8-object split) | F1: 47.02 -> 48.79 | Helped |
| Keep the pre-trained visual classifier fixed rather than tuning it on the caption subset (ablation) | NOC visual classifier · MSCOCO held-out objects (8-object split) | F1: 39.70 -> 48.79 | Hurt |
| Replace DCC's word2vec embeddings with GloVe (ablation) | DCC · MSCOCO held-out objects (8-object split) | F1 (%): 39.78 -> 38.04 | No clear effect |
| Change auxiliary-loss weights from (1, 1) to (1, 2) or (2, 1) (ablation) | NOC joint objective · MSCOCO validation set | F1 and METEOR | Hurt |
Closest papers by meaning
- Deep Compositional Captioning: Describing Novel Object Categories Without Paired Training Data (CVPR 2016 · similarity 0.93)
- Paraphrasing Is All You Need for Novel Object Captioning (NeurIPS 2022 · similarity 0.91)
- Beyond a Pre-Trained Object Detector: Cross-Modal Textual and Visual Context for Image Captioning (CVPR 2022 · similarity 0.91)
- Relational Distant Supervision for Image Captioning without Image-Text Pairs (AAAI 2024 · similarity 0.88)
- BACON: Improving Clarity of Image Captions via Bag-of-Concept Graphs (CVPR 2025 · similarity 0.88)
- 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.