Research map / Image generation / Super-resolution
Pixel Recursive Super Resolution
Ryan Dahl, Mohammad Norouzi, Jonathon Shlens
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 |
|---|---|---|---|
| use pixel recursive probabilistic model instead of L2 regression ResNet | super resolution · CelebA | % Fooled: 4.0 ± 0.2 -> 11.0±0.1 | Improved |
| use pixel recursive probabilistic model instead of L2 regression ResNet | super resolution · LSUN Bedrooms | % Fooled: 2.1 ± 0.1 -> 27.9±0.3 | Improved |
| use pixel recursive model instead of GAN baseline | super resolution · CelebA | % Fooled: 8.5 ± 0.2 -> 11.0±0.1 | Improved |
| lower sampling temperature from 1.0 to 0.8 (ablation) | pixel recursive SR · LSUN Bedrooms | % Fooled: 17.7 ± 0.4 -> 27.9±0.3 | Improved |
| lower sampling temperature from 1.0 to 0.8 (ablation) | pixel recursive SR · CelebA | % Fooled: 11.0±0.1 -> 10.2 ± 0.1 | Hurt |
| use greedy decoding (tau = 0) instead of sampling (ablation) | pixel recursive SR · CelebA / LSUN | perceptual quality | Hurt |
| use pixel recursive model instead of L2 regression | super resolution · LSUN Bedrooms | SSIM: 0.74 -> 0.58 | Hurt |
| use pixel recursive model instead of L2 regression | super resolution · CelebA | pSNR: 29.16 -> 29.09 | No clear effect |
| use pixel recursive model instead of GAN | super resolution · CelebA | Consistency (MSE): 0.029 -> 0.008 | Improved |
| add conditioning-network cross-entropy loss (Eq. 8) to the PixelCNN objective | pixel recursive SR · training | conditioning use | Improved |
| factor joint with PixelCNN autoregression instead of pixel-independent L2 or cross-entropy (ablation) | generative model · MNIST corners | sample mode | Improved |
Closest papers by meaning
- Image Super-Resolution Using Deep Convolutional Networks (TPAMI 2016 · similarity 0.90)
- Image Formation Model Guided Deep Image Super-Resolution (AAAI 2020 · similarity 0.89)
- Deeply-Recursive Convolutional Network for Image Super-Resolution (CVPR 2016 · similarity 0.89)
- Accelerating Image Super-Resolution Networks with Pixel-Level Classification (ECCV 2024 · similarity 0.89)
- Accurate Image Super-Resolution Using Very Deep Convolutional Networks (CVPR 2016 · similarity 0.89)
- Deep Back-Projection Networks for Super-Resolution (CVPR 2018 · 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.
