Research map / Image generation / GANs
Dual Contrastive Loss and Attention for GANs
Ning Yu, Guilin Liu, Aysegul Dundar, Andrew Tao, Bryan Catanzaro, Larry S. Davis, Mario Fritz
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 |
|---|---|---|---|
| replace non-saturating loss with dual contrastive loss (ablation) | StyleGAN2 · FFHQ, Bedroom, Church, Horse, CLEVR | FID: FFHQ 4.86 -> 3.98; Bedroom 4.01 -> 3.86; Church 4.54 -> 3.73; Horse 3.91 -> 3.70; CLEVR 9.62 -> 6.06 | Improved |
| add dual contrastive loss to other backbones (ablation) | SNGAN · CLEVR | FID: 29.19 -> 18.23 | Improved |
| add dual contrastive loss to other backbones (ablation) | StyleGAN · CLEVR | FID: 12.43 -> 8.96 | Improved |
| replace non-saturating loss with dual contrastive loss: discriminator feature distinguishability (ablation) | StyleGAN2 · FFHQ | FDDF: 245. -> 377. (FDDF) | Improved |
| add SAN self-attention in the generator (30k subsets) (ablation) | StyleGAN2 · Bedroom | FID: 19.33 -> 16.36 | Improved |
| add DFN attention in the generator (ablation) | StyleGAN2 · Bedroom | FID: 19.33 -> 26.86 | Hurt |
| add self-attention in the generator (large-scale) | StyleGAN2 · FFHQ vs Bedroom | FID: FFHQ 4.86 -> 5.13; Bedroom 4.01 -> 3.48 | Mixed |
| add self-attention in the discriminator (ablation) | StyleGAN2 · CelebA, Animal Face | FID: CelebA 9.84 -> 10.49; Animal Face 36.55 -> 42.41 | Hurt |
| add reference-attention in the discriminator (30k subsets) (ablation) | StyleGAN2 · Bedroom | FID: 19.33 -> 8.32 | Improved |
| add reference-attention in the discriminator at larger data size (100K) (ablation) | StyleGAN2 · CelebA 128x128, 100K images | FID: 5.61 -> 6.86 | Hurt |
| add reference-attention in the discriminator at 1K images (ablation) | StyleGAN2 · Bedroom 128x128, 1K images | FID: 230.40 -> 79.81 | Improved |
| combine reference-attention discriminator with dual contrastive loss (ablation) | StyleGAN2 w/ self-attn-G ref-attn-D · Bedroom (30k) | FID: 8.32 -> 12.84 | Hurt |
Closest papers by meaning
- Your Local GAN: Designing Two Dimensional Local Attention Mechanisms for Generative Models (CVPR 2020 · similarity 0.90)
- Projected GANs Converge Faster (NeurIPS 2021 · similarity 0.89)
- Are GANs Created Equal? A Large-Scale Study (NeurIPS 2018 · similarity 0.89)
- Self-Attention Generative Adversarial Networks (ICML 2019 · similarity 0.89)
- MCGAN: Enhancing GAN Training with Regression-Based Generator Loss (AAAI 2025 · 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.
