class Waifu2x(gpuid=0, model="models-cunet", tta_mode=False, num_threads=1, scale=2, noise=0, tilesize=0)
waifu2x-ncnn-vulkan class which can do image super resolution.
gpuid: int
the id of the gpu device to use. -1 for cpu mode.
model: str
the name or the path to the model
tta_mode: bool
whether to enable tta mode or not
num_threads: int
the number of threads in upscaling
default 1
scale: int
scale level, 1 = no scaling, 2 = upscale 2x
value: float. default: 2
noise: int
denoise level, large value means strong denoise effect, -1 = no effect
value: -1/0/1/2/3. default: -1
tilesize: int
tile size, use smaller value to reduce GPU memory usage, default selects automatically
0 for automatically setting the size. default: 0
Waifu2x.set_params(self, scale=2, noise=-1, tilesize=0)
set parameters for waifu2x object.
This function will be called during the object initialization, so usually no need to call it except for parameters changes after initialization.
scale: int
scale ratio.
value: 1/2. default: 2
noise: int
denoise level.
value: -1/0/1/2/3. default: -1
tilesize: int
tile size.
0 for automatically setting the size. default: 0
Waifu2x.load(parampath: str = "", modelpath: str = "")
Load models from given paths. Use Waifu2x.model if one or all of the parameters are not given.
This function will be called during the object initialization, so usually no need to call it except for setting a different Waifu2x.model after initialization.
parampath: str
the path to model params. usually ended with ".param".
modelpath: str
the path to model bin. usually ended with ".bin"
value: -1/0/1/2/3. default: -1
Waifu2x.process(self, im: PIL.Image)
Process the incoming PIL.Image
im: PIL.Image
the image object to process
The result PIL.Image object.
Waifu2x.gpuid
The id of gpu this Waifu2x Object is using.
Waifu2x.model
The model name or path this object is going to use. Waifu2x.load() should be called manually after updating this property.
Waifu2x.scale
The result scale ratio. It is different to the self._raw_w2xobj.scale. Waifu2x.scale controls the result scale size while self._raw_w2xobj.scale controls the scale ratio at each raw image process method call.
( Waifu2x._process(im) is the raw image process call. A upscaling task is done by repeatedly calling 2 times super-resolution)
Waifu2x._raw_w2xobj
The raw binding object of the original Waifu2x class. All the processing parameters are actually passed to this object eventually.
It is not recommend to operate on this object directly since there are many important parameter settings like tilesize setting have already been included in Python Waifu2x class.