57 lines
1.8 KiB
Python
57 lines
1.8 KiB
Python
|
#code originally taken from: https://github.com/ChenyangSi/FreeU (under MIT License)
|
||
|
|
|
||
|
|
import torch
|
||
|
|
|
||
|
|
|
||
|
|
def Fourier_filter(x, threshold, scale):
|
||
|
|
# FFT
|
||
|
|
x_freq = torch.fft.fftn(x.float(), dim=(-2, -1))
|
||
|
|
x_freq = torch.fft.fftshift(x_freq, dim=(-2, -1))
|
||
|
|
|
||
|
|
B, C, H, W = x_freq.shape
|
||
|
|
mask = torch.ones((B, C, H, W), device=x.device)
|
||
|
|
|
||
|
|
crow, ccol = H // 2, W //2
|
||
|
|
mask[..., crow - threshold:crow + threshold, ccol - threshold:ccol + threshold] = scale
|
||
|
|
x_freq = x_freq * mask
|
||
|
|
|
||
|
|
# IFFT
|
||
|
|
x_freq = torch.fft.ifftshift(x_freq, dim=(-2, -1))
|
||
|
|
x_filtered = torch.fft.ifftn(x_freq, dim=(-2, -1)).real
|
||
|
|
|
||
|
|
return x_filtered.to(x.dtype)
|
||
|
|
|
||
|
|
|
||
|
|
class FreeU:
|
||
|
|
@classmethod
|
||
|
|
def INPUT_TYPES(s):
|
||
|
|
return {"required": { "model": ("MODEL",),
|
||
|
|
"b1": ("FLOAT", {"default": 1.1, "min": 0.0, "max": 10.0, "step": 0.01}),
|
||
|
|
"b2": ("FLOAT", {"default": 1.2, "min": 0.0, "max": 10.0, "step": 0.01}),
|
||
|
|
"s1": ("FLOAT", {"default": 0.9, "min": 0.0, "max": 10.0, "step": 0.01}),
|
||
|
|
"s2": ("FLOAT", {"default": 0.2, "min": 0.0, "max": 10.0, "step": 0.01}),
|
||
|
|
}}
|
||
|
|
RETURN_TYPES = ("MODEL",)
|
||
|
|
FUNCTION = "patch"
|
||
|
|
|
||
|
|
CATEGORY = "_for_testing"
|
||
|
|
|
||
|
|
def patch(self, model, b1, b2, s1, s2):
|
||
|
|
def output_block_patch(h, hsp, transformer_options):
|
||
|
|
if h.shape[1] == 1280:
|
||
|
|
h[:,:640] = h[:,:640] * b1
|
||
|
|
hsp = Fourier_filter(hsp, threshold=1, scale=s1)
|
||
|
|
if h.shape[1] == 640:
|
||
|
|
h[:,:320] = h[:,:320] * b2
|
||
|
|
hsp = Fourier_filter(hsp, threshold=1, scale=s2)
|
||
|
|
return h, hsp
|
||
|
|
|
||
|
|
m = model.clone()
|
||
|
|
m.set_model_output_block_patch(output_block_patch)
|
||
|
|
return (m, )
|
||
|
|
|
||
|
|
|
||
|
|
NODE_CLASS_MAPPINGS = {
|
||
|
|
"FreeU": FreeU,
|
||
|
|
}
|