Research map / Image generation / Text-to-image diffusion
Reinforcing an Image Caption Generator Using Off-Line Human Feedback
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 |
|---|---|---|---|
| Use off-policy policy gradients with true human ratings instead of the baseline training alone | Image caption generator · T2 test | Goodness (Average): 66.23±0.60% -> 68.42±0.61%; 3.19% | Helped |
| Use off-policy policy gradients with true human ratings instead of the baseline training alone | Image caption generator · T2 test | Informativeness: 7.45±1.06% | Helped |
| Use off-policy policy gradients with true human ratings instead of the baseline training alone | Image caption generator · T2 test | Correctness: 5.90±0.80% | Helped |
| Use off-policy policy gradients with true human ratings instead of the baseline training alone | Image caption generator · T2 test | Fluency: 1.69±0.31% | Helped |
| Add positively rated captions at threshold 0.5 through fine-tuning | Baseline caption generator · T2 test | Goodness (Average): 66.23±0.60% -> 66.68±0.61%; 0.45% | No clear effect |
| Add positively rated captions at threshold 0.7 through fine-tuning | Baseline caption generator · T2 test | Goodness (Average): 66.23±0.60% -> 65.83±0.62%; -0.40% | No clear effect |
| Use on-policy policy gradients with an estimated rating function | Baseline caption generator · T2 test | Goodness (Average): 66.23±0.60% -> 65.97±0.61%; -0.26% | No clear effect |
| Increase the policy-gradient weight α for OffPG (ablation) | OffPG · T2 test, 200-image subset | Informativeness | Helped |
| Increase the policy-gradient weight α for OnPG (ablation) | OnPG · T2 test, 200-image subset | Informativeness | Hurt |
| Set α to 100 for OnPG (ablation) | OnPG · T2 test, 200-image subset | Caption diversity: α=100 | Hurt |
Closest papers by meaning
- CapRL: Stimulating Dense Image Caption Capabilities via Reinforcement Learning (ICLR 2026 · similarity 0.87)
- Rich Human Feedback for Text-to-Image Generation (CVPR 2024 · similarity 0.86)
- BalCapRL : A Balanced Framework for RL-Based MLLM Image Captioning (NeurIPS 2026 · similarity 0.85)
- Humor in AI: Massive Scale Crowd-Sourced Preferences and Benchmarks for Cartoon Captioning (NeurIPS 2024 · similarity 0.85)
- RePIC: Reinforced Post-Training for Personalizing Multi-Modal Language Models (NeurIPS 2025 · similarity 0.84)
- HIVE: Harnessing Human Feedback for Instructional Visual Editing (CVPR 2024 · similarity 0.84)
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.