NS_Upsampler

This is a super resolution application for ns gaming. It's designed for enhance the resolution from 1080p to 2160p. Or enhance the quality in theory(by model training data).

To use this, you should have a capture card hardware to project the picture to PC.

NS -> Dock -> Capture card -> PC -> Monitor

The first version available here: https://github.com/zhangchaosd/ns_upsampler/releases/tag/alpha

System requirments:

GPU: Intel GPU

TODO: NVIDIA GPUs

1. Train a Super-Resolution model

This step is to get the SRNet.onnx file. You can just use the file in the repo and go to Step 2.

1.1 Prepareing training data

• Use capture card and Obs Studio(or other software) to record raw 1080p gaming videos, put these videos to SRNet/raw_videos.

• Run python SRNet/prepare_data.py to extract frames from videos.

1.2 Training

You can change the hyper-parameters or the model as you like. In this repo, it's a very simple small model to ensure the low infer latency.

python SRNet/train.py

1.3 Test model (optional)

2. Compile the NS_Upsampler application

TODO

Download OpenCV and OpenVINO sdk, put them in the folder like this:

Visual Studio 2022

Open NS_SuperResolution/NS_SuperResolution.sln and build.

Remember copy OpenCV and OpenVINO dlls to the exe path to run.