Research map / Image generation / Editing and style transfer
GramGAN: Deep 3D Texture Synthesis From 2D Exemplars
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 |
|---|---|---|---|
| combine discriminator Gram style loss with the WGAN term (ablation) | GramGAN on wood · wood | SIFID × 10^5: 0.29 -> 0.14; 0.12 -> 0.14 | Mixed |
| combine discriminator Gram style loss with the WGAN term (ablation) | GramGAN on granite · granite | SIFID × 10^5: 0.43 -> 0.21; 0.28 -> 0.21 | Helped |
| replace the discriminator feature extractor with VGG for the style loss (ablation) | GramGAN · wood and granite | SIFID × 10^5: 0.14 -> 0.19; 0.21 -> 0.31 | Hurt |
| inject noise at the input layer rather than in hidden layers (ablation) | GramGAN · wood and granite | SIFID × 10^5: 0.14 -> 0.28; 0.21 -> 0.34 | Hurt |
| remove the style loss and use vanilla WGAN only (ablation) | GramGAN · wood and granite | SIFID × 10^5: 0.14 -> 0.29; 0.21 -> 0.28 | Mixed |
| remove the WGAN term and optimize only GramGAN style loss (ablation) | GramGAN · wood and granite | SIFID × 10^5: 0.14 -> 0.12; 0.21 -> 0.43 | Mixed |
| use GramGAN rather than Henzler et al. | single-exemplar synthesis · wood and granite | SIFID × 10^5: 0.67 -> 0.14; 0.53 -> 0.21 | Helped |
| use GramGAN rather than Henzler et al. | conditional texture-space synthesis · stone texture space, 22 unseen reference patches | SIFID × 10^5: 2.60 -> 2.47 | Helped |
| use GramGAN rather than Henzler et al. | conditional texture-space synthesis · stone texture space, 22 unseen reference patches | ALL: -319.02 -> -317.23 | Helped |
| use GramGAN rather than Henzler et al. | conditional texture-space synthesis · stone texture space, 22 unseen reference patches | rank from study: 2.28 -> 1.85 | Helped |
| fine-tune selected generator parameters using discriminator Gram style loss (ablation) | GramGAN · unseen texture exemplar | SIFID × 10^5: 2.04 -> 1.46 | Helped |
| use VGG instead of discriminator features during extrapolation (ablation) | GramGAN · unseen texture exemplar | SIFID × 10^5: 1.46 -> 1.79 | Hurt |
Closest papers by meaning
- Diversified Texture Synthesis With Feed-Forward Networks (CVPR 2017 · similarity 0.91)
- TextureGAN: Controlling Deep Image Synthesis With Texture Patches (CVPR 2018 · similarity 0.91)
- Texture Networks: Feed-forward Synthesis of Textures and Stylized Images (ICML 2016 · similarity 0.91)
- Learning a Neural 3D Texture Space From 2D Exemplars (CVPR 2020 · similarity 0.90)
- Fast Texture Synthesis via Pseudo Optimizer (CVPR 2020 · similarity 0.90)
- Paint-it: Text-to-Texture Synthesis via Deep Convolutional Texture Map Optimization and Physically-Based Rendering (CVPR 2024 · similarity 0.87)
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.