Research map / Vision-language and multimodal / Human and embodied modeling
Generating Diverse and Descriptive Image Captions Using Visual Paraphrases
Lixin Liu, Jiajun Tang, Xiaojun Wan, Zongming Guo
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 the first decoding stage (two-step) instead of training only on second-stage captions (ablation) | Ours (Tdiv, 0.1) · MS COCO test | Dist-3: 6957 -> 7810 | Improved |
| add the first decoding stage (two-step) instead of training only on second-stage captions (ablation) | Ours (Tdiv, 0.1) · MS COCO test | CIDEr: 102.6 -> 105.4 | Improved |
| add the first decoding stage over pretrain-then-finetune on second-stage captions (ablation) | Ours (Tdiv, 0.1) · MS COCO test | SPICE: 20.3 -> 21.0 | Improved |
| add the first decoding stage over pretrain-then-finetune on second-stage captions (ablation) | Ours (Tdiv, 0.1) · MS COCO test | R@1: 23.6 -> 25.0 | Improved |
| increase epsilon from 0.1 to 0.3 in TF-IDF diversity pair selection | Ours (Tdiv) · MS COCO test | Dist-3: 7810 -> 10053 | Improved |
| increase epsilon from 0.1 to 0.3 in TF-IDF diversity pair selection | Ours (Tdiv) · MS COCO test | CIDEr: 105.4 -> 86.9 | Hurt |
| increase epsilon from 0.1 to 0.3 in TF-IDF diversity pair selection | Ours (Tdiv) · MS COCO human evaluation | Diversity: 3.54 -> 4.06 | Improved |
| select pairs by sentence length instead of Tdiv | Ours (len, 0) · MS COCO test | CIDEr: 109.8 -> 69.7 | Hurt |
| select pairs by Yngve syntactic complexity | Ours (Yngve, 0) · MS COCO test | R@1: 19.2 -> 22.6 | Improved |
| select pairs by image retrieval rank | Ours (IR, 2) · MS COCO test | R@1: 19.2 -> 25.3 | Improved |
| select pairs by TF-IDF diversity (epsilon 0.1) over Attention baseline | Ours (Tdiv, 0.1) · MS COCO test | SPICE: 19.9 -> 21.0 | Improved |
| CIDEr optimization with reinforcement learning | CIDEr-RL · MS COCO test | Dist-3: 4972 -> 3694 | Hurt |
Closest papers by meaning
- From Show to Tell: A Survey on Deep Learning-Based Image Captioning (TPAMI 2022 · similarity 0.91)
- Speaking the Same Language: Matching Machine to Human Captions by Adversarial Training (ICCV 2017 · similarity 0.90)
- Generating More Pertinent Captions by Leveraging Semantics and Style on Multi-Source Datasets (IJCV 2023 · similarity 0.90)
- CAPability: A Comprehensive Visual Caption Benchmark for Evaluating Both Correctness and Thoroughness (NeurIPS 2025 · similarity 0.89)
- A Picture is Worth More Than 77 Text Tokens: Evaluating CLIP-Style Models on Dense Captions (CVPR 2024 · similarity 0.88)
- Enhancing Descriptive Captions with Visual Attributes for Multimodal Perception (CVPR 2026 · similarity 0.88)
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.
