Research map / Image generation / GANs
PacGAN: The power of two samples in generative adversarial networks
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 |
|---|---|---|---|
| Pack multiple samples at the discriminator input | GAN · 2D-ring | Modes: 6.3$\pm$0.5 -> 7.9$\pm$0.1 | Helped |
| Increase packing degree from 2 to 4 (ablation) | PacGAN on 2D-ring · 2D-ring | Modes: 7.9$\pm$0.1 -> 7.8$\pm$0.1 | No clear effect |
| Pack multiple samples at the discriminator input | GAN · 2D-grid | Modes: 17.3$\pm$0.8 -> 24.8$\pm$0.2 | Helped |
| Increase packing degree from 2 to 4 (ablation) | PacGAN on 2D-grid · 2D-grid | Modes: 23.8$\pm$0.7 -> 24.8$\pm$0.2 | Helped |
| Pack samples at the discriminator input | DCGAN · Stacked MNIST | Modes: 78.9$\pm$6.46 -> 1000.0$\pm$0.00 | Helped |
| Vary packing degree from 2 to 4 (ablation) | PacDCGAN on Stacked MNIST · Stacked MNIST | Modes: 1000.0$\pm$0.00 -> 1000.0$\pm$0.00 | No clear effect |
| Pack 2 samples at the discriminator input | DCGAN · CelebA, discriminator size $d^2$ | Probability of collision: $1$ -> $0.33$ | Helped |
| Pack 2 samples at the discriminator input | DCGAN · CelebA, discriminator size $4\,{d}^2$ | Probability of collision: $0.42$ -> $0$ | Helped |
| Pack 2 samples at the discriminator input | DCGAN · CelebA, discriminator size $16\,{d}^2$ | Probability of collision: $0.86$ -> $0$ | Helped |
| Pack 2 samples at the discriminator input | DCGAN · CelebA, discriminator size $25\,{d}^2$ | Probability of collision: $0.65$ -> $0.17$ | Helped |
| Pack samples at the discriminator input | WGAN · Stacked MNIST | Modes: 314.3$\pm$38.54 -> 965.7$\pm$19.07 | Helped |
| Increase packing degree from 2 to 4 (ablation) | PacWGAN on Stacked MNIST · Stacked MNIST | Modes: 927.6$\pm$22.19 -> 965.7$\pm$19.07 | Helped |
Closest papers by meaning
- Dual Discriminator Generative Adversarial Nets (NeurIPS 2017 · similarity 0.90)
- MGAN: Training Generative Adversarial Nets with Multiple Generators (ICLR 2018 · similarity 0.89)
- UniGAN: Reducing Mode Collapse in GANs using a Uniform Generator (NeurIPS 2022 · similarity 0.89)
- Mixed batches and symmetric discriminators for GAN training (ICML 2018 · similarity 0.89)
- Collaborative Sampling in Generative Adversarial Networks (AAAI 2020 · similarity 0.88)
- MCGAN: Enhancing GAN Training with Regression-Based Generator Loss (AAAI 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.