Research map / Image generation / GANs
A Framework of Composite Functional Gradient Methods for Generative Adversarial Models
Rie Johnson, Tong Zhang
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 |
|---|---|---|---|
| use xICFG instead of best baseline GAN/WGANgp | convolutional generators · MNIST | inception score: 9.57 -> 9.78 | Improved |
| use xICFG instead of best baseline GAN/WGANgp | convolutional generators · SVHN | inception score: 8.33 -> 8.72 | Improved |
| use xICFG instead of best baseline | fully-connected generators · SVHN | inception score: 7.81 -> 8.16 | Improved |
| use xICFG instead of best baseline | ResNet generators · LSUN BR+LR (64x64) | inception score: 1.56 -> 1.64 | Improved |
| use xICFG instead of best baseline | convolutional generators · LSUN T+B (64x64) | inception score: 1.76 -> 1.80 | Mixed |
| reduce number of ICFG iterations T from 25 to 5 (SVHN), 10 (MNIST), 15 (LSUN) (ablation) | xICFG · MNIST, SVHN, LSUN | inception score | No clear effect |
| use final approximator as generator instead of the full composite generator (ablation) | xICFG · MNIST, SVHN, LSUN | inception score | Hurt |
| remove batch normalization from all methods (ablation) | convolutional generators · image generation | inception score | No clear effect |
Closest papers by meaning
- Composite Functional Gradient Learning of Generative Adversarial Models (ICML 2018 · similarity 0.99)
- TRAINING GENERATIVE ADVERSARIAL NETWORKS VIA PRIMAL-DUAL SUBGRADIENT METHODS: A LAGRANGIAN PERSPECTIVE ON GAN (ICLR 2018 · similarity 0.91)
- On Stabilizing Generative Adversarial Training With Noise (CVPR 2019 · similarity 0.89)
- Optimizing Latent Distributions for Non-Adversarial Generative Networks (TPAMI 2020 · similarity 0.89)
- Towards Principled Methods for Training Generative Adversarial Networks (ICLR 2017 · similarity 0.89)
- Unrolled Generative Adversarial Networks (ICLR 2017 · similarity 0.89)
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.
