Research map / Image generation / Text-to-image diffusion
AttnGAN: Fine-Grained Text to Image Generation With Attentional Generative Adversarial Networks
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 the proposed DAMSM loss instead of using the prior text encoder replacement (ablation) | AttnGAN1 · CUB test | inception score: 3.98 -> 4.35 | Helped |
| add the proposed DAMSM loss instead of using the prior text encoder replacement (ablation) | AttnGAN1 · CUB test | R-precision: 10.37% -> 58.65% | Helped |
| increase DAMSM loss weight λ from 0.1 to 5 (ablation) | AttnGAN1 · CUB test | inception score: 4.19 -> 4.35 | Helped |
| increase DAMSM loss weight λ from 0.1 to 5 (ablation) | AttnGAN1 · CUB test | R-precision: 16.55% -> 58.65% | Helped |
| increase DAMSM loss weight λ from 5 to 10 (ablation) | AttnGAN1 · CUB test | inception score: 4.35 -> 4.29 | Hurt |
| increase DAMSM loss weight λ from 5 to 10 (ablation) | AttnGAN1 · CUB test | R-precision: 58.65% -> 63.87% | Helped |
| stack two attention models and use three generators instead of one attention model and two generators (ablation) | AttnGAN · CUB test | inception score: 4.35 -> 4.36 | Helped |
| fine-tune the pretrained DAMSM with the whole network (ablation) | AttnGAN · text-to-image synthesis | performance | No clear effect |
| apply DAMSM loss to outputs at all resolutions instead of only the last generator output (ablation) | AttnGAN · text-to-image synthesis | performance | No clear effect |
Closest papers by meaning
- Controllable Text-to-Image Generation (NeurIPS 2019 · similarity 0.88)
- DF-GAN: A Simple and Effective Baseline for Text-to-Image Synthesis (CVPR 2022 · similarity 0.88)
- StyleGAN-T: Unlocking the Power of GANs for Fast Large-Scale Text-to-Image Synthesis (ICML 2023 · similarity 0.87)
- Adversarial Text to Continuous Image Generation (CVPR 2024 · similarity 0.87)
- Text to Image Generation With Semantic-Spatial Aware GAN (CVPR 2022 · similarity 0.86)
- InstantPainting: Expanding GANs for Efficient Text-Conditioned Image Generation Platform (AAAI 2025 · 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.