Awesome

Demos

Demonstration implementations of neural network image processing algorithms. Everything is written from scratch in C# and has no dependencies on third-party libraries. Windows XP and higher, .NET Framework 4.0 and higher are required to run. Windows 7 and .NET Framework 4.5 are recommended. Any bit depth.

Important!

• All the presented implementations are inferior in performance to the corresponding author's code. All materials presented in this repository are intended solely for reference and are aimed at enabling neural networks to be run "out of the box" on home PCs, rather than for regular use. However, despite this, we are constantly working on optimization ;)
• The programs presented are extremely resource-intensive. Despite the theoretical possibility of running them on old PCs, you should not expect high code speed even on modern machines.
• Commercial use of some of the algorithms implemented here is prohibited by the authors. Carefully study the license in the official repositories!
• Detailed instructions for launching implementations can be found in the file: MANUAL.md.

News

• [03.11.2021] The process of refactoring and optimizing codes has been launched. The repository is being reorganized.
• [03.10.2020] The entire Colorization block is now completely in English! Descriptions, comments, and the program interface are now in the same language.
• [02.08.2020] Fixed some bugs in "Arbitrary Style Transfer with Style-Attentional Networks".
• [29.07.2020] Updated weights for "Arbitrary Style Transfer with Style-Attentional Networks", updated and added new examples. The implementation of this work is now an alpha version.
• [23.05.2020] Added implementation of the "ChromaGAN: Adversarial Picture Colorization with Semantic Class Distribution". Added FAQ section (Frequently Asked Questions).
• [08.03.2020] Added implementation of the "Let there be Color!: Joint End-to-End Learning of Global and Local Image Priors for Automatic Image Colorization with Simultaneous Classification". Fixed minor bugs.
• [09.12.2019] Added the implementation of "Photo - Realistic Single Image Super - Resolution Using a Generative Adversarial Network"(SRGAN). Fixed an error displaying demo images to implementations.
• [09.11.2019] Correction of minor errors.
• [16.10.2019] Optimized sparse convolution (dilated conv2d) in Colorful Image Colorization.
• [16.10.2019] Added the repository icon.
• [30.06.2019] Updated implementation of "Avatar - Net: Multi-scale Zero-shot Style Transfer by Feature Decoration": weights converted to Half format, demo images converted to jpeg format.
• [30.06.2019] Updated implementation of "Perceptual Losses for Real-Time Style Transfer and Super - Resolution": weights converted to Half format, demo images converted to jpeg format.
• [30.06.2019] Updated implementation of "Arbitrary Style Transfer in Real - time with Adaptive Instance Normalization": weights converted to Half format, demo images converted to jpeg format.
• [30.06.2019] Updated examples for TensorZoom.
• [30.06.2019] Updated implementation of "Universal Style Transfer via Feature Transforms": weights converted to Half format, duplicate weights removed, demo images converted to jpeg format.
• [11.06.2019] Added instructions for launching the codes presented in this repository.
• [11.06.2019] The implementation of the "Arbitrary Style Transfer with Style - Attentive Networks" paper is placed.
• [02.06.2019] Updated implementation of the Gatys's Algorithm: weights are converted to Half format, and demo images are converted to jpeg format.
• [21.05.2019] Updated examples for the Gatys's Algorithm.
• [20.05.2019] Several new examples of the Gatys's Algorithm have been published.
• [10.05.2019] An implementation of the "Fast Patch - based Style Transfer of Arbitrary Style" is posted.
• [05.05.2019] Implemented the Waifu2x neural network (only super - resolution for photos).
• [03.05.2019] Implemented the work of Justin Johnson's "Perceptual Losses for Real - Time Style Transfer and Super - Resolution".
• [28.03.2019] A new section "Colorization" has been opened and the implementation of "Colorful Image Colorization" by Richard Sheng has been placed; the implementation of AvatarNet has been fixed and the examples to It have been updated.
• [14.03.2019] Optimized the ConvTranspose2D layer in TensorZoom.
• [11.03.2019] An implementation of the work "Avatar-Net: Multi - scale Zero - shot Style Transfer by Feature Decoration" by Lukas Sheng is posted.