Research map / Label-efficient and robust learning / Recognition and long-tail
Training GANs with Stronger Augmentations via Contrastive Discriminator
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 |
|---|---|---|---|
| Use ContraD with SimCLR augmentations instead of bCR with HFlip and translation | SNDCGAN · CIFAR-10 | FID: 14.0 -> 10.9 | Helped |
| Use ContraD with SimCLR augmentations instead of bCR with HFlip and translation | SNDCGAN · CIFAR-100 | FID: 19.2 -> 15.2 | Helped |
| Use ContraD with SimCLR augmentations instead of bCR | SNResNet-18 discriminator GAN · CIFAR-10 | FID: 22.8 -> 9.86 | Helped |
| Use ContraD with SimCLR augmentations instead of bCR | SNResNet-18 discriminator GAN · CIFAR-100 | FID: 28.2 -> 15.0 | Helped |
| Add ContraD instead of bCR | SNDCGAN · CelebA-HQ-128 | FID: 19.5 -> 17.1 | Helped |
| Use ContraD instead of ADA | StyleGAN2 · AFHQ-Cat | FID: 3.55 -> 3.82 | Hurt |
| Use ContraD instead of ADA | StyleGAN2 · AFHQ-Dog and AFHQ-Wild | FID: AFHQ-Dog: 7.40 -> 7.16; AFHQ-Wild: 3.05 -> 2.54 | Helped |
| Add stop-gradient before the discriminator head (ablation) | ContraD · CIFAR-10 | FID: SNDCGAN: 11.6 -> 11.1; SNResNet-18: 28.0 -> 10.6 | Helped |
| Use a 2-layer MLP instead of a linear discriminator head (ablation) | ContraD · CIFAR-10 | FID: SNDCGAN: 185 -> 11.1; SNResNet-18: 274 -> 10.6 | Helped |
| Add the real-sample contrastive loss L⁺con (ablation) | ContraD · CIFAR-10 | FID: SNDCGAN: 11.9 -> 11.1; SNResNet-18: 182 -> 10.6 | Helped |
| Add the fake-sample contrastive loss L⁻con (ablation) | ContraD · CIFAR-10 | FID: SNDCGAN: 210 -> 11.1; SNResNet-18: 232 -> 10.6 | Helped |
| Replace SimCLR augmentations with HFlip and Trans (ablation) | ContraD · CIFAR-10 | FID: 10.9 -> 13.7 | Hurt |
Closest papers by meaning
- Augmentation-Aware Self-Supervision for Data-Efficient GAN Training (NeurIPS 2023 · similarity 0.90)
- Training Generative Adversarial Networks with Limited Data (NeurIPS 2020 · similarity 0.90)
- Improving GANs with A Dynamic Discriminator (NeurIPS 2022 · similarity 0.89)
- Improving Generalization and Stability of Generative Adversarial Networks (ICLR 2019 · similarity 0.89)
- Discriminative Forests Improve Generative Diversity for Generative Adversarial Networks (AAAI 2024 · similarity 0.88)
- Direct Discriminative Optimization: Your Likelihood-Based Visual Generative Model is Secretly a GAN Discriminator (ICML 2025 · 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.