Research map / Image generation / GANs
CVAE-GAN: Fine-Grained Image Generation Through Asymmetric Training
Jianmin Bao, Dong Chen, Fang Wen, Houqiang Li, Gang Hua
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 mean feature matching to conditional GAN (FM-CGAN vs CGAN) | CGAN · FaceScrub generated faces | top-1 accuracy of generated samples: 61.97% -> 79.76% | Improved |
| add encoder with pair-wise feature matching and pixel loss (CVAE-GAN vs FM-CGAN) | FM-CGAN · FaceScrub generated faces | top-1 accuracy of generated samples: 79.76% -> 97.78% | Improved |
| add mean feature matching to CGAN | CGAN · FaceScrub generated faces | realism (Inception Score): 15.79 -> 19.40 | Improved |
| add encoder pair-wise matching to FM-CGAN | FM-CGAN · FaceScrub generated faces | realism (Inception Score): 19.40 -> 19.03 | No clear effect |
| use CVAE-GAN instead of CVAE | CVAE · FaceScrub generated faces | top-1 accuracy of generated samples: 8.09% -> 97.78% | Improved |
| add adversarial feature loss L_G(D) (ablation) | CVAE-GAN · FaceScrub, CUB-200, Flower (qualitative) | image sharpness | Improved |
| add pixel-level reconstruction loss L_G(img) (ablation) | CVAE-GAN · FaceScrub, CUB-200, Flower (qualitative) | image detail | Improved |
| add classification feature loss L_G(C) (ablation) | CVAE-GAN · FaceScrub, CUB-200, Flower (qualitative) | category information | Improved |
| augment training with generated images of existing identities | face recognition model · LFW | verification accuracy: 91.87% -> 92.77% | Improved |
| augment training with generated images of 5k new mixed identities | face recognition model · LFW | verification accuracy: 91.87% -> 92.98% | Improved |
Later papers that compared with this method
3 later accepted papers report CVAE-GAN as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.
- Ph-GAN: Physics-Inspired GAN for Generating SAR Images Under Limited Data (ICCV 2025 · better on 3 of 3 shared results)
- Learning towards Selective Data Augmentation for Dialogue Generation (AAAI 2023 · better on 2 of 2 shared results)
- Cooperative Training of Fast Thinking Initializer and Slow Thinking Solver for Conditional Learning (TPAMI 2021 · better on 5 of 5 shared results)
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
- Semi-Supervised Single-Stage Controllable GANs for Conditional Fine-Grained Image Generation (ICCV 2021 · similarity 0.90)
- Autoencoding beyond pixels using a learned similarity metric (ICML 2016 · similarity 0.88)
- Dual Projection Generative Adversarial Networks for Conditional Image Generation (ICCV 2021 · similarity 0.88)
- Ensembling With Deep Generative Views (CVPR 2021 · similarity 0.88)
- Cluster-Guided Image Synthesis With Unconditional Models (CVPR 2022 · similarity 0.88)
- Adversarial Latent Autoencoders (CVPR 2020 · similarity 0.88)
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.
