Research map / Probabilistic and scientific modeling / Variational and generative sampling
Explicitly Minimizing the Blur Error of Variational Autoencoders
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 |
|---|---|---|---|
| Add a reconstruction loss that explicitly weights estimated blur errors | VAE · CelebA 64×64 | FID_Gen: 0.0575 -> 0.0536 | Helped |
| Add a reconstruction loss that explicitly weights estimated blur errors | VAE · CelebA-HQ 256×256 | FID_Gen: 0.0982 -> 0.0938 | Helped |
| Add a reconstruction loss that explicitly weights estimated blur errors | VAE · HCP | SSIM: 0.5922$\pm$2.1$e^{-1}$ -> 0.5209$\pm$1.9$e^{-1}$ | Hurt |
| Add a reconstruction loss that explicitly weights estimated blur errors | VAE · CIFAR10 | FID_Gen: 0.0618 -> 0.0829 | Hurt |
| Use a z-dependent covariance with large ε rather than small ε (ablation) | Proposed VAE loss · CelebA 64×64 | PSNR: 19.51 -> 23.21 | Helped |
| Use a z-dependent covariance with large ε rather than a fixed Gaussian covariance (ablation) | Proposed VAE loss · CelebA 64×64 | PSNR: 22.56 -> 23.21 | Helped |
| Set Wiener-filter constant C to 0.1 instead of 0.025 (ablation) | Proposed VAE loss · CelebA 64×64 | PSNR: 23.21 -> 22.87 | Hurt |
| Use C=0.005 rather than C=0.025 (ablation) | Proposed VAE loss · CelebA 64×64 | PSNR: 23.21 -> 23.21 | No clear effect |
| Use a per-image kernel generator Gγ(z) rather than a fixed kernel (ablation) | Proposed VAE loss · CelebA 64×64 | PSNR: 22.56 -> 23.21 | Helped |
Closest papers by meaning
- From Variational to Deterministic Autoencoders (ICLR 2020 · similarity 0.91)
- A Contrastive Learning Approach for Training Variational Autoencoder Priors (NeurIPS 2021 · similarity 0.91)
- MAE: Mutual Posterior-Divergence Regularization for Variational AutoEncoders (ICLR 2019 · similarity 0.90)
- Variational Autoencoder with Implicit Optimal Priors (AAAI 2019 · similarity 0.89)
- Unscented Autoencoder (ICML 2023 · similarity 0.89)
- Be a Goldfish: Forgetting Bad Conditioning in Sparse Linear Regression via Variational Autoencoders (ICML 2025 · similarity 0.89)
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.