Research map / Vision-language and multimodal / Human and embodied modeling
Length-Controllable Image Captioning
Chaorui Deng, Ning Ding, Mingkui Tan, Qi Wu
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 4-level length level embedding | VLP · MS COCO Karpathy test, normal length | CIDEr-D: 116.9 -> 118.7 | Improved |
| add 5-level length level embedding | AoANet · MS COCO Karpathy test | SPICE: 21.3 -> 23.0 | Improved |
| add 4-level length level embedding | LaBERT · MS COCO test, level 2 | CIDEr-D: 116.8 -> 118.2 | Improved |
| add 4-level length level embedding | LaBERT · MS COCO test, level 3 | SPICE: 21.7 -> 22.3 | Improved |
| generate captions at longest length levels | 4-Level VLP · MS COCO test | CIDEr-D: 118.7 -> 40.0 | Hurt |
| reduce refine steps from 25 to 15 (ablation) | 4-Level LaBERT · MS COCO test, level 4 | SPICE: 22.2 -> 22.0 | No clear effect |
| reduce refine steps from 25 to 12 (ablation) | 4-Level LaBERT · MS COCO test, level 4 | SPICE: 22.2 -> 21.9 | No clear effect |
| add global update rule (ablation) | 4-Level LaBERT · MS COCO test, level 2 | CIDEr-D: 116.6 -> 118.2 | Improved |
| add [EOS] decay (ablation) | 4-Level LaBERT · MS COCO test, level 2 | CIDEr-D: 116.0 -> 118.2 | Improved |
| use 4-level length embedding instead of beam search | AoANet · MS COCO test | SelfCIDEr: 0.590 -> 0.689 | Improved |
| use 4-level length embedding instead of beam search | VLP · MS COCO test | SelfCIDEr: 0.623 -> 0.762 | Improved |
| use non-autoregressive LaBERT instead of 4-level VLP | LaBERT · MS COCO test | SelfCIDEr: 0.762 -> 0.841 | Improved |
Closest papers by meaning
- Image Captioning With Controllable and Adaptive Length Levels (TPAMI 2023 · similarity 0.95)
- Guiding Image Captioning Models Toward More Specific Captions (ICCV 2023 · similarity 0.90)
- Show, Control and Tell: A Framework for Generating Controllable and Grounded Captions (CVPR 2019 · similarity 0.90)
- Controllable Image Captioning via Prompting (AAAI 2023 · similarity 0.89)
- Painting with Words: Elevating Detailed Image Captioning with Benchmark and Alignment Learning (ICLR 2025 · similarity 0.87)
- CapsFusion: Rethinking Image-Text Data at Scale (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.
