Research map / Image generation / GANs
SONA: Learning Conditional, Unconditional, and Matching-Aware Discriminator
Yuhta Takida, Satoshi Hayakawa, Takashi Shibuya, Masaaki Imaizumi, Naoki Murata, Bac Nguyen, Toshimitsu Uesaka, Chieh-Hsin Lai, Yuki Mitsufuji
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 orthogonal projection in the alignment term (ablation) | SONA · CIFAR10 | FID: 7.51 -> 6.29 | Improved |
| add the mismatch loss BT-mm (ablation) | SONA · CIFAR10 | IS: 9.08 -> 9.54 | Improved |
| add both orthogonal projection and mismatch loss (ablation) | SONA · CIFAR10 | FID: 7.51 -> 5.65 | Improved |
| add adaptive weighting of the three objectives (ablation) | SONA · CIFAR10 | FID: 7.09 -> 5.65 | Improved |
| remove mismatching loss (ablation) | SONA · 2D mixture of Gaussians | class separation | Hurt |
| use SONA instead of PD-GAN on mixture of Gaussians with many classes | conditional GAN discriminator · MoG, N >= 30 | number of failure cases | Improved |
| replace projection discriminator with SONA on BigGAN | PD-GAN · CIFAR10 | FID: 4.60 -> 4.24 | Improved |
| replace projection discriminator with SONA | PD-GAN · ImageNet 128 batch 256 | FID: 29.76 -> 13.17 | Improved |
| combine SONA with DiffAug | SONA · TinyImageNet | FID: 16.33 -> 7.76 | Improved |
| apply SONA to GALIP discriminator with frozen CLIP text embedding | GALIP · COCO | FID: 5.30 -> 4.70 | Improved |
| apply SONA to GALIP | GALIP · CUB | CLIP Score: 0.3310 -> 0.3342 | No clear effect |
| compare against ReACGAN on StyleGAN2 CIFAR10 FID | SONA · CIFAR10 StyleGAN2 | FID: 3.39 -> 3.38 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- PD-GAN — PD-GAN: Probabilistic Diverse GAN for Image Inpainting (CVPR 2021) · better on 7 of 7 shared results
- ContraGAN — ContraGAN: Contrastive Learning for Conditional Image Generation (NeurIPS 2020) · 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
- Dual Projection Generative Adversarial Networks for Conditional Image Generation (ICCV 2021 · similarity 0.87)
- ContraGAN: Contrastive Learning for Conditional Image Generation (NeurIPS 2020 · similarity 0.86)
- Turning Waste into Wealth: Leveraging Low-Quality Samples for Enhancing Continuous Conditional Generative Adversarial Networks (AAAI 2024 · similarity 0.85)
- CASUAL: Conditional Support Alignment for Domain Adaptation with Label Shift (AAAI 2025 · similarity 0.84)
- Direct Discriminative Optimization: Your Likelihood-Based Visual Generative Model is Secretly a GAN Discriminator (ICML 2025 · similarity 0.84)
- Discriminative Probing and Tuning for Text-to-Image Generation (CVPR 2024 · 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.
