Research map / Reinforcement learning / Visual and representation RL
Deep Reinforcement Learning-Based Image Captioning With Embedding Reward
Check 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 policy and value networks with embedding-driven actor-critic training and lookahead inference | image captioning · MS COCO | CIDEr: 0.872 -> 0.937 | Helped |
| add image-sentence embedding to rank final candidates (ablation) | SL image captioning · MS COCO | CIDEr: 0.872 -> 0.888 | Helped |
| add pretrained raw value network and lookahead inference (ablation) | SL image captioning · MS COCO | CIDEr: 0.872 -> 0.916 | Helped |
| add embedding-driven actor-critic reinforcement learning after SL-RawVN (ablation) | SL-RawVN · MS COCO | CIDEr: 0.916 -> 0.937 | Helped |
| use policy hidden state as value-network input instead of independent raw image and sentence inputs (ablation) | Full-model · MS COCO | CIDEr: 0.937 -> 0.69 | Hurt |
| use policy hidden state and policy image feature as value-network input instead of independent raw inputs (ablation) | Full-model · MS COCO | CIDEr: 0.937 -> 0.701 | Hurt |
| combine policy and value guidance with λ=0.4 rather than only value guidance (λ=0) (ablation) | lookahead inference · MS COCO | CIDEr: 0.8 -> 0.937 | Helped |
| combine policy and value guidance with λ=0.4 rather than only policy guidance (λ=1) (ablation) | lookahead inference · MS COCO | CIDEr: 0.879 -> 0.937 | Helped |
| increase beam size from 5 to 100 (ablation) | SL beam-search baseline · MS COCO | CIDEr: 0.876 -> 0.849 | Hurt |
| increase beam size from 5 to 100 (ablation) | lookahead inference · MS COCO | CIDEr: 0.934 -> 0.92 | No clear effect |
Closest papers by meaning
- CapRL: Stimulating Dense Image Caption Capabilities via Reinforcement Learning (ICLR 2026 · similarity 0.91)
- Temporal-Difference Learning With Sampling Baseline for Image Captioning (AAAI 2018 · similarity 0.90)
- BalCapRL : A Balanced Framework for RL-Based MLLM Image Captioning (NeurIPS 2026 · similarity 0.89)
- RePIC: Reinforced Post-Training for Personalizing Multi-Modal Language Models (NeurIPS 2025 · similarity 0.86)
- ViCrit: A Verifiable Reinforcement Learning Proxy Task for Visual Perception in VLMs (NeurIPS 2025 · similarity 0.85)
- Learning by Correction: Efficient Tuning Task for Zero-Shot Generative Vision-Language Reasoning (CVPR 2024 · similarity 0.85)
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.