Research map / Image generation / GANs
LatentAugment: Data Augmentation via Guided Manipulation of GAN’s Latent Space
Lorenzo Tronchin, Minh H. Vu, Paolo Soda, Tommy Löfstedt
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 the three diversity losses (pixel, perceptual, latent) (ablation) | LatentAugment (MAE objective) · MRI-to-CT test set | MAE: 41.99 -> 39.39 | Improved |
| add the latent diversity loss (ablation) | LatentAugment · MRI-to-CT test set | MAE: 41.29 -> 39.39 | Improved |
| add the perceptual diversity loss (ablation) | LatentAugment · MRI-to-CT test set | MAE: 40.46 -> 39.39 | Improved |
| add the pixel diversity loss (ablation) | LatentAugment · MRI-to-CT test set | MAE: 39.92 -> 39.39 | Improved |
| add the fidelity term alpha_f (ablation) | LatentAugment · MRI-to-CT test set | MAE: 40.05 -> 39.39 | Improved |
| use LatentAugment instead of no augmentation | Pix2Pix · MRI-to-CT test set | MAE: 45.64 -> 39.32 | Improved |
| use LatentAugment instead of standard SG2 sampling DA | Pix2Pix · MRI-to-CT test set | MAE: 42.26 -> 39.39 | Improved |
| use standard geometric DA instead of none | Pix2Pix · MRI-to-CT test set | MAE: 45.64 -> 44.73 | No clear effect |
| use standard SG2 DA instead of none | Pix2Pix · MRI-to-CT test set | MAE: 45.64 -> 42.26 | Improved |
| optimize hyper-parameters on F1 score instead of MAE | LatentAugment · MRI-to-CT test set | LPIPS: 6.33e-2 -> 6.10e-2 | Improved |
| add rotation/scale/aniso/translation augmentation to multimodal SG2 training | StyleGAN2 multimodal · CT FID | FID: 14.97 -> 9.10 | Improved |
| add xflip and int-style augmentations on top of geometric ones in SG2 training (ablation) | StyleGAN2 multimodal · MRI FID | FID: 7.12 -> 11.35 | Hurt |
Closest papers by meaning
- Differentiable Augmentation for Data-Efficient GAN Training (NeurIPS 2020 · similarity 0.89)
- Complexity Matters: Rethinking the Latent Space for Generative Modeling (NeurIPS 2023 · similarity 0.89)
- Distribution Augmentation for Generative Modeling (ICML 2020 · similarity 0.88)
- A Data-Augmentation Is Worth A Thousand Samples: Analytical Moments And Sampling-Free Training (NeurIPS 2022 · similarity 0.88)
- Diverse Rare Sample Generation with Pretrained GANs (AAAI 2025 · similarity 0.87)
- Effective Data Augmentation With Diffusion Models (ICLR 2024 · similarity 0.86)
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.
