Research map / Probabilistic and scientific modeling / Variational and generative sampling
PixelCNN++: Improving the PixelCNN with Discretized Logistic Mixture Likelihood and Other Modifications
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 |
|---|---|---|---|
| discretized logistic mixture likelihood instead of continuous mixture with uniform dequantization (ablation) | PixelCNN++ · CIFAR-10 | test bits per sub-pixel: 2.92 vs 3.11 | Helped |
| 256-way softmax output instead of discretized logistic mixture (ablation) | PixelCNN++ · CIFAR-10 | training convergence and speed | Hurt |
| remove short-cut connections (ablation) | PixelCNN++ · CIFAR-10 | trainability | Hurt |
| remove dropout regularization (ablation) | PixelCNN++ · CIFAR-10 | test bits per sub-pixel: final test above 6.0 vs 2.92 | Hurt |
| add NIN blocks to small-receptive-field PixelCNN (ablation) | Small PixelCNN · CIFAR-10 | test bits per sub-pixel: 3.11 to 3.09 (11x5); 3.07 to 3.04 (15x8) | Helped |
| add autoregressive channel connections to small-receptive-field PixelCNN (ablation) | Small PixelCNN · CIFAR-10 | test bits per sub-pixel: 3.11 to 3.07 (11x5); 3.07 to 3.03 (15x8) | Helped |
| downsampling instead of dilated convolutions (with shortcuts) | PixelCNN++ · CIFAR-10 | test bits per sub-pixel | No clear effect |
| class-conditioning on labels | PixelCNN++ · CIFAR-10 | test bits per sub-pixel: 2.94 vs 2.92 | Hurt |
Closest papers by meaning
- PixelCNN Models with Auxiliary Variables for Natural Image Modeling (ICML 2017 · similarity 0.88)
- Pixel Recurrent Neural Networks (ICML 2016 · similarity 0.88)
- Conditional Image Generation with PixelCNN Decoders (NeurIPS 2016 · similarity 0.87)
- Approximation and Learning with Deep Convolutional Models: a Kernel Perspective (ICLR 2022 · similarity 0.87)
- PixelDiT2: Representation-Grounded Pixel Diffusion Transformers (NeurIPS 2026 · similarity 0.86)
- Direct Discriminative Optimization: Your Likelihood-Based Visual Generative Model is Secretly a GAN Discriminator (ICML 2025 · 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.