Special thanks to Sky Kim. Visit Sky's GitHub.
Original Model: naoto0804 / pytorch-AdaIN
Double-click the Basic Scene (Assets-Scenes-Basic), and you can add your own textures and run the AdaIN model.
Double-click the TTS Scene (Assets-Scenes-TTS). This scene is designed to use the camera's target texture as the content input, and you can render the output to the screen. However, please note that there are some indexing issues because PyTorch usually uses CHW instead of CWH.
- Known Issues:
- Output resolution is fixed at 512x512. If you want a different resolution, it would be better to export the ONNX file's shape like this, for example: FullHD (1080x1920). You can edit PyTorch codes or reshape inputs & output tensors using ShallowReshape.
- The camera needs to be rotated typically ±180 following the Z-axis.
For Korean readers, I have published articles on my personal blog. Please read the articles below.
#1 How To Install Sentis on your Unity project + From Barracuda to Sentis migration guide
#2 Unity Barracuda vs. Unity Sentis benchmarks
#3 Let's apply StyleTransfer (AdaIN) to New3DURPSample
🚨 Please note that I do not hold any copyrights for the sources used in this project. 🚨