Description

waifu2x filter for VapourSynth, based on waifu2x-caffe.

Note

The folder models must be located in the same folder as Waifu2x-caffe.dll.

Requires cudnn64_7.dll (cuDNN v7.6.5) to be in the search path. Due to the license of cuDNN, I can't distribute the required dll file. You have to register and download yourself at https://developer.nvidia.com/cudnn.

Usage

caffe.Waifu2x(clip clip[, int noise=0, int scale=2, int block_w=128, int block_h=block_w, int model=6, bint cudnn=True, int processor=0, bint tta=False, int batch=1])

• clip: Clip to process. Only planar format with float sample type of 32 bit depth is supported.

• noise: Noise reduction level.

  • -1 = none
  • 0 = low
  • 1 = medium
  • 2 = high
  • 3 = highest

• scale: Upscaling factor. Must be a power of 2. Set to 1 for no upscaling.

• block_w: The horizontal block size for dividing the image during processing. Smaller value results in lower VRAM usage, while larger value may not necessarily give faster speed. The optimal value may vary according to different graphics card and image size. It's recommended that the source's width is divisible by the block size.

• block_h: The same as block_w but for vertical. It's recommended that the source's height is divisible by the block size.

• model: Sets which model to use. Only anime_style_art is Y model (luma only), the others are RGB models.

  • 0 = anime_style_art (for 2D artwork)
  • 1 = anime_style_art_rgb (for 2D artwork)
  • 2 = photo (for photo and anime)
  • 3 = upconv_7_anime_style_art_rgb (has faster speed than anime_style_art_rgb, with equal or better quality)
  • 4 = upconv_7_photo (has faster speed than photo, with equal or better quality)
  • 5 = upresnet10 (has better quality than upconv_7_anime_style_art_rgb). Note that the result will change if the block size is different. The recommended block size of upresnet10 is 38. Use a larger batch size to compensate the slowness due to small block size.
  • 6 = cunet (has the best quality for 2D artwork among the bundled models). Note that the result will change if the block size is different. The block size of cunet must be divisible by 4.

• cudnn: When set to true, it uses cuDNN for processing. When set to false, CUDA will be used instead.

• processor: Sets which GPU device to use. The device number begins with 0. The default device will be selected if a nonexistent device is specified.

• tta: Whether to use TTA(Test-Time Augmentation) mode. It increases PSNR by 0.15 or so, but 8 times slower.

• batch: The batch size for concurrent processing of blocks.

Compilation

Requires customized Caffe library. To build Caffe, you must have all the dependencies: CUDA Toolkit 10, cuDNN 7, OpenBLAS, Boost, protobuf, glog, and gflags. Additionally, waifu2x-caffe library requires OpenCV 4. The defaults in Makefile.config should work, but adjust the relevant lines if not (such as CUDA_DIR). Then just type make all -j4 to compile Caffe.

There are options caffe_includedir, caffe_libdir, cudaincludedir, and cudalibdir settable in Meson.

meson build
ninja -C build
ninja -C build install