Research map / Vision-language and multimodal / Human and embodied modeling
Look Back and Predict Forward in Image Captioning
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 |
|---|---|---|---|
| Add previous-step attention result to current attention generation | Up-Down image captioner · MSCOCO Karpathy test split | CIDEr: 113.5 -> 114.3 | Helped |
| Predict two successive words in one time step and combine their probabilities | Up-Down image captioner · MSCOCO Karpathy test split | CIDEr: 113.5 -> 115.7 | Helped |
| Combine previous-step attention with two-word forward prediction | Up-Down image captioner · MSCOCO Karpathy test split | CIDEr: 113.5 -> 116.4 | Helped |
| Add previous-step attention result to current attention generation | Up-Down image captioner with CIDEr optimization · MSCOCO Karpathy test split | CIDEr: 120.1 -> 124.4 | Helped |
| Predict two successive words in one time step and combine their probabilities | Up-Down image captioner with CIDEr optimization · MSCOCO Karpathy test split | CIDEr: 120.1 -> 126.9 | Helped |
| Combine previous-step attention with two-word forward prediction | Up-Down image captioner with CIDEr optimization · MSCOCO Karpathy test split | CIDEr: 120.1 -> 127.6 | Helped |
| Set λ to 0.5 instead of λ=0, combining the two predicted sequences (ablation) | PF · MSCOCO Karpathy test split, cross-entropy loss | CIDEr: 114.4 -> 115.7 | Helped |
| Set λ to 0.5 instead of λ=0, combining the two predicted sequences (ablation) | LBPF · MSCOCO Karpathy test split, cross-entropy loss | CIDEr: 114.8 -> 116.4 | Helped |
| Set λ to 0.3 instead of λ=0 for PF (ablation) | PF with CIDEr optimization · MSCOCO Karpathy test split | CIDEr: 126.5 -> 126.9 | Helped |
| Set λ to 0.5 instead of λ=0 for LBPF (ablation) | LBPF with CIDEr optimization · MSCOCO Karpathy test split | CIDEr: 127.1 -> 127.6 | Helped |
| Increase λ from 0.5 to 0.9 (ablation) | PF · MSCOCO Karpathy validation dataset | CIDEr: 113.8 -> 113.5 | Hurt |
| Increase λ from 0.5 to 0.9 (ablation) | LBPF · MSCOCO Karpathy validation dataset | CIDEr: 114.3 -> 113.8 | Hurt |
Closest papers by meaning
- Knowing When to Look: Adaptive Attention via a Visual Sentinel for Image Captioning (CVPR 2017 · similarity 0.90)
- Deliberate Attention Networks for Image Captioning (AAAI 2019 · similarity 0.89)
- Visual Attention Never Fades: Selective Progressive Attention ReCalibration for Detailed Image Captioning in Multimodal Large Language Models (ICML 2025 · similarity 0.87)
- Painting with Words: Elevating Detailed Image Captioning with Benchmark and Alignment Learning (ICLR 2025 · similarity 0.86)
- Rethinking Causal Mask Attention for Vision-Language Inference (ICLR 2026 · similarity 0.85)
- Noise-Aware Image Captioning with Progressively Exploring Mismatched Words (AAAI 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.