Research map / Probabilistic and scientific modeling / Variational and generative sampling
PixelVAE: A Latent Variable Model for Natural Images
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 PixelCNN-based autoregressive decoder to a VAE (PixelVAE) versus standard VAE | VAE decoder · binarized MNIST | NLL test upper bound | Helped |
| Use only one autoregressive PixelCNN layer versus none in PixelVAE (ablation) | PixelVAE · binarized MNIST | KL term / NLL bound | Helped |
| Replace upsampling convolutional latent conditioning with direct linear conditioning of each PixelCNN layer (ablation) | Gated PixelVAE · binarized MNIST | NLL test upper bound: 79.66 vs 80.02 upper bound | Helped |
| Use fewer than 10 PixelCNN layers in PixelVAE rather than PixelCNN of same depth (ablation) | PixelVAE · binarized MNIST | NLL | Helped |
| Replace PixelCNN layers at the higher latent level with a diagonal Gaussian decoder in hierarchical PixelVAE (ablation) | Hierarchical PixelVAE · LSUN bedrooms | log-likelihood | Hurt |
| Hierarchical PixelVAE versus PixelRNN on 64x64 ImageNet | Hierarchical PixelVAE · 64x64 ImageNet | NLL (bits/dim): 3.66 vs 3.63 | Hurt |
Closest papers by meaning
- NVAE: A Deep Hierarchical Variational Autoencoder (NeurIPS 2020 · similarity 0.91)
- Variational Lossy Autoencoder (ICLR 2017 · similarity 0.90)
- Very Deep VAEs Generalize Autoregressive Models and Can Outperform Them on Images (ICLR 2021 · similarity 0.90)
- PixelCNN Models with Auxiliary Variables for Natural Image Modeling (ICML 2017 · similarity 0.89)
- Soft-IntroVAE: Analyzing and Improving the Introspective Variational Autoencoder (CVPR 2021 · similarity 0.89)
- EQ-VAE: Equivariance Regularized Latent Space for Improved Generative Image Modeling (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.