Research map / Vision-language and multimodal / Human and embodied modeling
CaptionSmiths: Flexibly Controlling Language Pattern in Image Captioning
Kuniaki Saito, Donghyun Kim, Kwanyong Park, Atsushi Hashimoto, Yoshitaka Ushiku
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 descriptiveness control to length control (ablation) | CaptionSmiths · MSCOCO (Short) | CIDEr: 89.0 -> 98.8 | Improved |
| add uniqueness control to length control (ablation) | CaptionSmiths · MSCOCO (Short) | CIDEr: 89.0 -> 95.1 | Improved |
| add uniqueness control on top of length and descriptiveness (ablation) | CaptionSmiths · MSCOCO (Short) | CIDEr: 98.8 -> 104.8 | Improved |
| add descriptiveness control on top of length and uniqueness (ablation) | CaptionSmiths · MSCOCO (Short) | CIDEr: 95.1 -> 104.8 | Improved |
| replace discrete cluster conditioning (5 groups) with continuous interpolation (ablation) | CaptionSmiths · COCO (trained on COCO only) | CIDEr: 110 -> 112 | Improved |
| replace discrete cluster conditioning (100 groups) with continuous interpolation (ablation) | CaptionSmiths · COCO (trained on COCO only) | CIDEr: 103 -> 112 | Improved |
| replace discrete conditioning (5 groups) with continuous interpolation (ablation) | CaptionSmiths · COCO (trained on COCO only) | length mismatch: 13.8 -> 1.6 | Improved |
| increase discrete group count from 5 to 100 (ablation) | discrete conditioning · COCO (trained on COCO only) | length mismatch and CIDEr: length mismatch 13.8 -> 9.7; CIDEr 110 -> 103 | Mixed |
| add descriptiveness and uniqueness conditions to length-only LC baseline | CaptionSmiths · COCO (trained on COCO only) | CIDEr: 103.6 -> 112.6 | Improved |
| add target-caption conditioning to LLaVA trained on mixed data | CaptionSmiths · MSCOCO (Short) | CIDEr: 13.2 -> 104.8 | Improved |
| use continuous sample-specific conditioning instead of dataset-aware Concap | CaptionSmiths · Docci (Long) | CIDEr: 8.3 -> 29.7 | Improved |
| use CaptionSmiths instead of Concap on COCO | CaptionSmiths · MSCOCO (Short) | BLEU@4: 11.5 -> 11.4 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- BLIP-3 — BLIP: Bootstrapping Language-Image Pre-training for Unified Vision-Language Understanding and Generation (ICML 2022) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Image Captioning With Controllable and Adaptive Length Levels (TPAMI 2023 · similarity 0.89)
- Length-Controllable Image Captioning (ECCV 2020 · similarity 0.88)
- Semantic and Expressive Variations in Image Captions Across Languages (CVPR 2025 · similarity 0.88)
- Guiding Image Captioning Models Toward More Specific Captions (ICCV 2023 · similarity 0.88)
- FlexCap: Describe Anything in Images in Controllable Detail (NeurIPS 2024 · similarity 0.87)
- Text as Partial Constraint: Core–Residual Alignment for Robust Vision–Language Learning (NeurIPS 2026 · 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.
