Classical Texture Synthesis and Beyond
summary
Compares four concrete algorithms: Image Quilting with minimum-cost seam and patch-size/overlap/tolerance knobs (FFT or kd-tree for nearest neighbors), Texture Optimization as MRF energy minimization with multi-resolution EM-like iterations, Texture Synthesis using CNNs via Gram matrices in VGG optimized from noise with L-BFGS, and Non-Stationary GANs that expand patches 2× using adversarial + L1 + style losses and per-texture training; shows which textures each method actually reproduces, runtimes, and typical failure modes.
AI-written summary · about
chapters (7)
- 0:00Texture synthesis intro
- 3:41Classical synthesis algorithms
- 11:46Image quilting parameters
- 17:27Neural synthesis with CNNs
- 20:58Non-stationary GAN synthesis
- 24:52Comparing synthesis methods
- 26:58Conclusion
related talks
10mKnitGIST: A Programming Synthesis Toolkit for Generating Functional Machine-Knitting Textures
17mKnitPicking Textures: Programming and Modifying Complex Knitted Textures for Machine and Hand ...
40mA Box of Chaos: The Generative Artist's Toolkit
30mVisual data abundant relevant labelled cheap Pick two
26mMaximizing Multimodal: Exploring the search frontier of text-to-image models
49mGraph Based Processing of Big Images
25mMixing art with Python- an introduction to Style Transfer
45mHow to Hack a Painting