- Windows 7 or up, 64-bit only
- An Nvidia GPU, the newer and the more memory, the better.
- Python 3 64-bit (x86-64) : https://www.python.org/downloads/release/python-366/
- Step 1: Install CUDA 9.2 + the patch. I believe you only need the CUDA and driver components: https://developer.nvidia.com/cuda-toolkit-archive
- Step 2: Go to the PyTorch installation site. Use “pip” as the package, “stable” for the build, and everything else should match your software versions. Open up the Windows command prompt (search “cmd” in the start menu) and paste in the commands. https://pytorch.org/get-started/locally/
- If it tells you “x is not a supported wheel on this platform”, make absolutely sure your options match and that your Python installation is 64-bit, or try different versions of Python or CUDA.
- While you’re there, paste in this (pip3, not pip): pip3 install numpy opencv-python
- Step 3: Download https://github.com/xinntao/ESRGAN along with the two pretrained models.
- Step 4: Follow the instructions in ESRGAN’s “Quick Test” section. You can skip step 1, beacause you already downloaded the program. I highly recommend running the network interpolation command, because the default, full monty SRGAN model can make images a bit noisy whereas PSNR is too blurry.
RAW Paste Data