Research map / Image generation / GANs
StackGAN++: Realistic Image Synthesis with Stacked Generative Adversarial Networks
Han Zhang, Tao Xu, Hongsheng Li, Shaoting Zhang, Xiaogang Wang, Xiaolei Huang, Dimitris Metaxas
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 Conditioning Augmentation (ablation) | StackGAN-v1 256x256 · CUB | Inception score: 3.31 -> 3.70 | Improved |
| add Conditioning Augmentation (ablation) | Stage-I GAN 256x256 · CUB | Inception score: 2.48 -> 3.02 | Improved |
| add Conditioning Augmentation (ablation) | Stage-I GAN 64x64 · CUB | Inception score: 2.66 -> 2.95 | Improved |
| add Stage-II refinement on top of Stage-I (ablation) | StackGAN-v1 · CUB | Inception score: 3.02 -> 3.70 | Improved |
| feed text again at Stage-II (ablation) | StackGAN-v1 256x256 · CUB | Inception score: 3.45 -> 3.70 | Improved |
| feed text again at Stage-II (ablation) | StackGAN-v1 128x128 · CUB | Inception score: 3.13 -> 3.35 | Improved |
| increase output resolution from 128x128 to 256x256 (ablation) | StackGAN-v1 · CUB | Inception score: 3.35 -> 3.70 | Improved |
| add intermediate multi-scale branches G1, G2 (ablation) | StackGAN-v2 · CUB | Inception score: 3.49 -> 4.04 | Improved |
| add multi-scale branches versus all-256 branches (ablation) | StackGAN-v2 · CUB | Inception score: 2.89 -> 4.04 | Improved |
| use multiple 256x256 discriminators with different initializations (ablation) | StackGAN-v2-G3 · CUB | Inception score: 3.49 -> 3.22 | Hurt |
| add joint conditional and unconditional discriminator (JCU) (ablation) | StackGAN-v2 · CUB | Inception score: 3.77 -> 4.04 | Improved |
| add color-consistency regularization (ablation) | StackGAN-v2 · ImageNet-dog | Inception score: 9.02 -> 9.55 | Improved |
Later papers that compared with this method
3 later accepted papers report StackGANv2 as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- OneGAN: Simultaneous Unsupervised Learning of Conditional Image Generation, Foreground Segmentation, and Fine-Grained Clustering (ECCV 2020 · better on 2 of 2 shared results)
- MixNMatch: Multifactor Disentanglement and Encoding for Conditional Image Generation (CVPR 2020 · better on 0 of 1 shared result)
- AttnGAN: Fine-Grained Text to Image Generation With Attentional Generative Adversarial Networks (CVPR 2018 · better on 1 of 1 shared result)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- StackGAN: Text to Photo-Realistic Image Synthesis With Stacked Generative Adversarial Networks (ICCV 2017 · similarity 0.95)
- MSG-GAN: Multi-Scale Gradients for Generative Adversarial Networks (CVPR 2020 · similarity 0.89)
- StudioGAN: A Taxonomy and Benchmark of GANs for Image Synthesis (TPAMI 2023 · similarity 0.89)
- DF-GAN: A Simple and Effective Baseline for Text-to-Image Synthesis (CVPR 2022 · similarity 0.89)
- Large Scale GAN Training for High Fidelity Natural Image Synthesis (ICLR 2019 · similarity 0.89)
- Adversarial Text to Continuous Image Generation (CVPR 2024 · similarity 0.87)
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.
