Spaces:
Running on Zero
Running on Zero
File size: 5,482 Bytes
d066167 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 | import cv2
import numpy as np
import torch
import torch.nn as nn
def postprocess(pred, thresh=0.18):
assert thresh <= 1.0 and thresh >= 0.0
pred = torch.amax(pred, 0)
pred[pred < thresh] = 0
pred -= 0.5
pred *= 2
return pred
class SketchKeras(nn.Module):
def __init__(self):
super(SketchKeras, self).__init__()
self.downblock_1 = nn.Sequential(
nn.ReflectionPad2d((1, 1, 1, 1)),
nn.Conv2d(1, 32, kernel_size=3, stride=1),
nn.BatchNorm2d(32, eps=1e-3, momentum=0),
nn.ReLU(),
)
self.downblock_2 = nn.Sequential(
nn.ReflectionPad2d((1, 1, 1, 1)),
nn.Conv2d(32, 64, kernel_size=4, stride=2),
nn.BatchNorm2d(64, eps=1e-3, momentum=0),
nn.ReLU(),
nn.ReflectionPad2d((1, 1, 1, 1)),
nn.Conv2d(64, 64, kernel_size=3, stride=1),
nn.BatchNorm2d(64, eps=1e-3, momentum=0),
nn.ReLU(),
)
self.downblock_3 = nn.Sequential(
nn.ReflectionPad2d((1, 1, 1, 1)),
nn.Conv2d(64, 128, kernel_size=4, stride=2),
nn.BatchNorm2d(128, eps=1e-3, momentum=0),
nn.ReLU(),
nn.ReflectionPad2d((1, 1, 1, 1)),
nn.Conv2d(128, 128, kernel_size=3, stride=1),
nn.BatchNorm2d(128, eps=1e-3, momentum=0),
nn.ReLU(),
)
self.downblock_4 = nn.Sequential(
nn.ReflectionPad2d((1, 1, 1, 1)),
nn.Conv2d(128, 256, kernel_size=4, stride=2),
nn.BatchNorm2d(256, eps=1e-3, momentum=0),
nn.ReLU(),
nn.ReflectionPad2d((1, 1, 1, 1)),
nn.Conv2d(256, 256, kernel_size=3, stride=1),
nn.BatchNorm2d(256, eps=1e-3, momentum=0),
nn.ReLU(),
)
self.downblock_5 = nn.Sequential(
nn.ReflectionPad2d((1, 1, 1, 1)),
nn.Conv2d(256, 512, kernel_size=4, stride=2),
nn.BatchNorm2d(512, eps=1e-3, momentum=0),
nn.ReLU(),
)
self.downblock_6 = nn.Sequential(
nn.ReflectionPad2d((1, 1, 1, 1)),
nn.Conv2d(512, 512, kernel_size=3, stride=1),
nn.BatchNorm2d(512, eps=1e-3, momentum=0),
nn.ReLU(),
)
self.upblock_1 = nn.Sequential(
nn.Upsample(scale_factor=2, mode="bicubic"),
nn.ReflectionPad2d((1, 2, 1, 2)),
nn.Conv2d(1024, 512, kernel_size=4, stride=1),
nn.BatchNorm2d(512, eps=1e-3, momentum=0),
nn.ReLU(),
nn.ReflectionPad2d((1, 1, 1, 1)),
nn.Conv2d(512, 256, kernel_size=3, stride=1),
nn.BatchNorm2d(256, eps=1e-3, momentum=0),
nn.ReLU(),
)
self.upblock_2 = nn.Sequential(
nn.Upsample(scale_factor=2, mode="bicubic"),
nn.ReflectionPad2d((1, 2, 1, 2)),
nn.Conv2d(512, 256, kernel_size=4, stride=1),
nn.BatchNorm2d(256, eps=1e-3, momentum=0),
nn.ReLU(),
nn.ReflectionPad2d((1, 1, 1, 1)),
nn.Conv2d(256, 128, kernel_size=3, stride=1),
nn.BatchNorm2d(128, eps=1e-3, momentum=0),
nn.ReLU(),
)
self.upblock_3 = nn.Sequential(
nn.Upsample(scale_factor=2, mode="bicubic"),
nn.ReflectionPad2d((1, 2, 1, 2)),
nn.Conv2d(256, 128, kernel_size=4, stride=1),
nn.BatchNorm2d(128, eps=1e-3, momentum=0),
nn.ReLU(),
nn.ReflectionPad2d((1, 1, 1, 1)),
nn.Conv2d(128, 64, kernel_size=3, stride=1),
nn.BatchNorm2d(64, eps=1e-3, momentum=0),
nn.ReLU(),
)
self.upblock_4 = nn.Sequential(
nn.Upsample(scale_factor=2, mode="bicubic"),
nn.ReflectionPad2d((1, 2, 1, 2)),
nn.Conv2d(128, 64, kernel_size=4, stride=1),
nn.BatchNorm2d(64, eps=1e-3, momentum=0),
nn.ReLU(),
nn.ReflectionPad2d((1, 1, 1, 1)),
nn.Conv2d(64, 32, kernel_size=3, stride=1),
nn.BatchNorm2d(32, eps=1e-3, momentum=0),
nn.ReLU(),
)
self.last_pad = nn.ReflectionPad2d((1, 1, 1, 1))
self.last_conv = nn.Conv2d(64, 1, kernel_size=3, stride=1)
def forward(self, x):
d1 = self.downblock_1(x)
d2 = self.downblock_2(d1)
d3 = self.downblock_3(d2)
d4 = self.downblock_4(d3)
d5 = self.downblock_5(d4)
d6 = self.downblock_6(d5)
u1 = torch.cat((d5, d6), dim=1)
u1 = self.upblock_1(u1)
u2 = torch.cat((d4, u1), dim=1)
u2 = self.upblock_2(u2)
u3 = torch.cat((d3, u2), dim=1)
u3 = self.upblock_3(u3)
u4 = torch.cat((d2, u3), dim=1)
u4 = self.upblock_4(u4)
u5 = torch.cat((d1, u4), dim=1)
out = self.last_conv(self.last_pad(u5))
return out
def proceed(self, img):
img = np.array(img)
blurred = cv2.GaussianBlur(img, (0, 0), 3)
img = img.astype(int) - blurred.astype(int)
img = img.astype(np.float32) / 127.5
img /= np.max(img)
img = torch.tensor(img).unsqueeze(0).permute(3, 0, 1, 2).cuda()
img = self(img)
img = postprocess(img, thresh=0.1).unsqueeze(1)
return img |