Research map / Image generation / Autoregressive and latent generation
Combining Markov Random Fields and Convolutional Neural Networks for Image Synthesis
Check 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 an MRF prior over middle-/upper-layer neural patches | CNN-MRF image synthesis · Photorealistic style transfer | visual plausibility | Helped |
| Add an MRF prior over neural patches instead of using Gram-matrix style matching | VGG-19 style transfer · Style transfer | local consistency | Helped |
| Add middle-layer neural patch matching at relu3_1 and relu4_1 instead of pixel or lower-/top-layer matching (ablation) | VGG-19 patch synthesis · Neural patch matching | match quality | Helped |
| Use relu5_1 patches instead of relu3_1 or relu4_1 patches (ablation) | VGG-19 patch synthesis · Neural patch matching | match quality | Hurt |
| Blend semantically related patches in middle-layer neural feature space instead of pixel space (ablation) | VGG-19 patch synthesis · Neural patch blending | blend quality | Helped |
| Remove the MRF regularizer (ablation) | CNN-MRF image synthesis · Style transfer | local consistency | Hurt |
| Use pixel-space patch optimization instead of neural-level MRF optimization (ablation) | CNN-MRF image synthesis · Image synthesis | synthesis quality | Hurt |
| Use the MRF prior when style and content have incompatible structure or perspective | CNN-MRF image synthesis · Style transfer with strong structural mismatch | content preservation | Hurt |
| Add an MRF style constraint rather than relying only on the more adaptable parametric style method | Image style transfer · Style transfer | style fidelity | Mixed |
Closest papers by meaning
- Neural Texture Synthesis With Guided Correspondence (CVPR 2023 · similarity 0.90)
- PixelNN: Example-based Image Synthesis (ICLR 2018 · similarity 0.87)
- Diversified Texture Synthesis With Feed-Forward Networks (CVPR 2017 · similarity 0.87)
- Texture Networks: Feed-forward Synthesis of Textures and Stylized Images (ICML 2016 · similarity 0.86)
- MarkovGen: Structured Prediction for Efficient Text-to-Image Generation (CVPR 2024 · similarity 0.85)
- Generative Photography: Scene-Consistent Camera Control for Realistic Text-to-Image Synthesis (CVPR 2025 · similarity 0.85)
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.