Research map / Image generation / GANs
Improving Image Captioning with Conditional Generative Adversarial Nets
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 |
|---|---|---|---|
| Add discriminator-based adversarial training | bottom-up+top-down SCST · MSCOCO Karpathy test | BLEU-4: 36.3 -> 38.3 | Helped |
| Add discriminator-based adversarial training | bottom-up+top-down SCST · MSCOCO Karpathy test | CIDEr: 120.1 -> 123.2 | Helped |
| Add discriminator-based adversarial training | bottom-up+top-down SCST · MSCOCO Karpathy test | SPICE: 21.4 -> 22.1 | Helped |
| Ensemble 4 CNN-GAN and 4 RNN-GAN models | bottom-up+top-down · MSCOCO Karpathy test | CIDEr: 123.2 -> 125.9 | Helped |
| Use an RNN rather than CNN discriminator | bottom-up+top-down adversarial training · MSCOCO Karpathy test | CIDEr: 123.2 -> 122.2 | Mixed |
| Add adversarial training and model ensembling | Up-Down · MSCOCO evaluation server c5 | CIDEr: 117.9 -> 123.1 | Helped |
| Increase the discriminator reward weight from lambda 0 to lambda 0.3 (ablation) | bottom-up+top-down CNN-GAN · MSCOCO Karpathy test | BLEU-4: 0.363 -> 0.383 | Helped |
| Use CIDEr-D rather than CIDEr as the language evaluator (ablation) | bottom-up+top-down CNN-GAN · MSCOCO Karpathy test | BLEU-4: 0.381 -> 0.383 | Helped |
| Use CIDEr-D rather than CIDEr as the language evaluator (ablation) | bottom-up+top-down CNN-GAN · MSCOCO Karpathy test | CIDEr: 1.248 -> 1.232 | Hurt |
| Increase discriminator updates from d=1 to d=5 (ablation) | bottom-up+top-down CNN-GAN · MSCOCO Karpathy test | BLEU-4: 0.383 -> 0.378 | Hurt |
| Increase generator updates from g=1 to g=5 (ablation) | bottom-up+top-down CNN-GAN · MSCOCO Karpathy test | BLEU-4: 0.383 -> 0.383 | No clear effect |
| Increase generator updates from g=1 to g=10 (ablation) | bottom-up+top-down CNN-GAN · MSCOCO Karpathy test | BLEU-4: 0.383 -> 0.381 | Hurt |
Closest papers by meaning
- Interactive Dual Generative Adversarial Networks for Image Captioning (AAAI 2020 · similarity 0.91)
- BalCapRL : A Balanced Framework for RL-Based MLLM Image Captioning (NeurIPS 2026 · similarity 0.88)
- CapRL: Stimulating Dense Image Caption Capabilities via Reinforcement Learning (ICLR 2026 · similarity 0.88)
- RePIC: Reinforced Post-Training for Personalizing Multi-Modal Language Models (NeurIPS 2025 · similarity 0.86)
- Learning by Correction: Efficient Tuning Task for Zero-Shot Generative Vision-Language Reasoning (CVPR 2024 · similarity 0.86)
- GMAIL: Generative Modality Alignment for generated Image Learning (ICML 2025 · 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.