Research map / Detection and segmentation / Open-set and few-shot detection
Partially-Supervised Image 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 |
|---|---|---|---|
| Train with image labels using PS3 rather than captions alone | Up-Down · COCO novel object validation, out-of-domain | CIDEr: 69.5 -> 94.3 | Helped |
| Use PS3 training instead of relying on constrained decoding with predicted labels | Up-Down · COCO novel object validation, out-of-domain | CIDEr: 74.8 -> 94.3 | Helped |
| Add constrained beam search decoding after PS3 training (ablation) | Up-Down with PS3 · COCO novel object validation, out-of-domain | CIDEr: 94.3 -> 92.5 | Hurt |
| Train with image labels using PS3 | Up-Down · COCO novel object validation, in-domain | CIDEr: 108.6 -> 101.2 | Hurt |
| Train jointly with labels for 25 Open Images animal classes | Up-Down · COCO Karpathy val | CIDEr: 112.3 -> 103.5 | Hurt |
| Add fixed word embeddings without constrained decoding (ablation) | Up-Down · COCO novel object validation, out-of-domain | CIDEr: 69.5 -> 66.4 | Hurt |
| Use both GloVe and dependency embeddings rather than GloVe alone with T2of3 decoding (ablation) | Up-Down · COCO novel object validation, out-of-domain | CIDEr: 80.2 -> 84.1 | Helped |
| Use both GloVe and dependency embeddings rather than GloVe alone with T3of3 decoding (ablation) | Up-Down · COCO novel object validation, out-of-domain | CIDEr: 80.1 -> 82.2 | Helped |
| Select three labels and require at least two to be mentioned, rather than the tested alternatives (ablation) | PS3 image-caption training · COCO novel object captioning | captioning scores | Helped |
Closest papers by meaning
- Beyond a Pre-Trained Object Detector: Cross-Modal Textual and Visual Context for Image Captioning (CVPR 2022 · similarity 0.91)
- Captioning Images With Diverse Objects (CVPR 2017 · similarity 0.90)
- Unsupervised Image Captioning (CVPR 2019 · similarity 0.90)
- Convolutional Image Captioning (CVPR 2018 · similarity 0.90)
- EVCap: Retrieval-Augmented Image Captioning with External Visual-Name Memory for Open-World Comprehension (CVPR 2024 · similarity 0.88)
- Relational Distant Supervision for Image Captioning without Image-Text Pairs (AAAI 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.