AI and Deep learning tools for Unity using CNTK
Note
This project was developed for Aalto University's Computational Intelligence in Games course material.
The development is stopped now, because we decided to use Tensorflowsharp with Unity MLAgent instead of CNTK for multiplatform support. The new project will be in public soon. Here.
Content
This rep contains some useful deep learning related tools implemented primarily using CNTK C# library. Current contents:
- Helper functions to build/train neural network layers. (https://docs.microsoft.com/en-us/cognitive-toolkit/)
- Layers definitions
- Simple neural network
- cGAN
- Universal Style Transfer(https://arxiv.org/pdf/1705.08086.pdf)
- Reinforcement Learning
- Proximal Policy Optimization(PPO)(https://arxiv.org/pdf/1707.06347.pdf)
- Deep Q-Learning(DQL)(https://arxiv.org/abs/1312.5602)
Platform and Installation
Currently it only works on Windows. If you need to use GPU for NN, you also need a proper Nvidia graphic card. Installation steps:
- Download the repo(Unity project)
- Download the zip that includes necessary dlls https://drive.google.com/open?id=1VWEiXJw3PSdeXfBrimPevdCWdbUyHd_0
- Put the dlls in correct places: (Adapted from https://github.com/aiunderstand/unity-cntk/blob/master/README.md.)
- Put those files/folders into any Plugins folder under /DeepLearningToolsForUnity/Assets.
- Cntk.Core.Managed-2.4.dll
- MathNet.Numerics.dll
- MathNet.Numerics.MKL.dll
- System.Drawing.dll
- Accord folder
- Copy the other dlls(not folders), and put them DIRECTLY under /DeepLearningToolsForUnity folder, or another place where Windows can find those dlls.
- Done.
Note that the file Assets/UnityCNTK/Tools/UniversalStyleTransfer/Data/UST_combined.bytes uses Git LFS, be sure you download it correctly (It should be larger than 100MB)
Documentation
Go to Wiki to see detailed documentaion.