Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -4,43 +4,44 @@ import torch.nn as nn
|
|
| 4 |
import gradio as gr
|
| 5 |
from PIL import Image
|
| 6 |
import torchvision.transforms as transforms
|
| 7 |
-
import os
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 8 |
|
| 9 |
-
# 🧠 Neural network layers
|
| 10 |
norm_layer = nn.InstanceNorm2d
|
| 11 |
|
| 12 |
-
# 🧱 Building block for the generator
|
| 13 |
class ResidualBlock(nn.Module):
|
| 14 |
def __init__(self, in_features):
|
| 15 |
super(ResidualBlock, self).__init__()
|
| 16 |
-
|
| 17 |
conv_block = [ nn.ReflectionPad2d(1),
|
| 18 |
nn.Conv2d(in_features, in_features, 3),
|
| 19 |
norm_layer(in_features),
|
| 20 |
nn.ReLU(inplace=True),
|
| 21 |
nn.ReflectionPad2d(1),
|
| 22 |
nn.Conv2d(in_features, in_features, 3),
|
| 23 |
-
norm_layer(in_features)
|
| 24 |
-
|
| 25 |
-
|
| 26 |
self.conv_block = nn.Sequential(*conv_block)
|
| 27 |
|
| 28 |
def forward(self, x):
|
| 29 |
return x + self.conv_block(x)
|
| 30 |
|
| 31 |
-
# 🎨 Generator model for creating line drawings
|
| 32 |
class Generator(nn.Module):
|
| 33 |
def __init__(self, input_nc, output_nc, n_residual_blocks=9, sigmoid=True):
|
| 34 |
super(Generator, self).__init__()
|
| 35 |
-
|
| 36 |
-
#
|
| 37 |
model0 = [ nn.ReflectionPad2d(3),
|
| 38 |
nn.Conv2d(input_nc, 64, 7),
|
| 39 |
norm_layer(64),
|
| 40 |
nn.ReLU(inplace=True) ]
|
| 41 |
self.model0 = nn.Sequential(*model0)
|
| 42 |
|
| 43 |
-
#
|
| 44 |
model1 = []
|
| 45 |
in_features = 64
|
| 46 |
out_features = in_features*2
|
|
@@ -52,13 +53,13 @@ class Generator(nn.Module):
|
|
| 52 |
out_features = in_features*2
|
| 53 |
self.model1 = nn.Sequential(*model1)
|
| 54 |
|
| 55 |
-
#
|
| 56 |
model2 = []
|
| 57 |
for _ in range(n_residual_blocks):
|
| 58 |
model2 += [ResidualBlock(in_features)]
|
| 59 |
self.model2 = nn.Sequential(*model2)
|
| 60 |
|
| 61 |
-
#
|
| 62 |
model3 = []
|
| 63 |
out_features = in_features//2
|
| 64 |
for _ in range(2):
|
|
@@ -69,85 +70,216 @@ class Generator(nn.Module):
|
|
| 69 |
out_features = in_features//2
|
| 70 |
self.model3 = nn.Sequential(*model3)
|
| 71 |
|
| 72 |
-
#
|
| 73 |
model4 = [ nn.ReflectionPad2d(3),
|
| 74 |
-
|
| 75 |
if sigmoid:
|
| 76 |
model4 += [nn.Sigmoid()]
|
| 77 |
-
|
| 78 |
self.model4 = nn.Sequential(*model4)
|
| 79 |
|
| 80 |
-
def forward(self, x
|
| 81 |
out = self.model0(x)
|
| 82 |
out = self.model1(out)
|
| 83 |
out = self.model2(out)
|
| 84 |
out = self.model3(out)
|
| 85 |
out = self.model4(out)
|
| 86 |
-
|
| 87 |
return out
|
| 88 |
|
| 89 |
-
#
|
| 90 |
-
|
| 91 |
-
model1
|
| 92 |
-
|
| 93 |
-
|
| 94 |
-
|
| 95 |
-
|
| 96 |
-
model2.
|
| 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 |
iface.launch()
|
|
|
|
| 4 |
import gradio as gr
|
| 5 |
from PIL import Image
|
| 6 |
import torchvision.transforms as transforms
|
| 7 |
+
import os
|
| 8 |
+
from huggingface_hub import hf_hub_download
|
| 9 |
+
import torch.nn.functional as F
|
| 10 |
+
|
| 11 |
+
# Check for CUDA availability but fallback to CPU
|
| 12 |
+
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
|
| 13 |
|
|
|
|
| 14 |
norm_layer = nn.InstanceNorm2d
|
| 15 |
|
|
|
|
| 16 |
class ResidualBlock(nn.Module):
|
| 17 |
def __init__(self, in_features):
|
| 18 |
super(ResidualBlock, self).__init__()
|
| 19 |
+
|
| 20 |
conv_block = [ nn.ReflectionPad2d(1),
|
| 21 |
nn.Conv2d(in_features, in_features, 3),
|
| 22 |
norm_layer(in_features),
|
| 23 |
nn.ReLU(inplace=True),
|
| 24 |
nn.ReflectionPad2d(1),
|
| 25 |
nn.Conv2d(in_features, in_features, 3),
|
| 26 |
+
norm_layer(in_features) ]
|
| 27 |
+
|
|
|
|
| 28 |
self.conv_block = nn.Sequential(*conv_block)
|
| 29 |
|
| 30 |
def forward(self, x):
|
| 31 |
return x + self.conv_block(x)
|
| 32 |
|
|
|
|
| 33 |
class Generator(nn.Module):
|
| 34 |
def __init__(self, input_nc, output_nc, n_residual_blocks=9, sigmoid=True):
|
| 35 |
super(Generator, self).__init__()
|
| 36 |
+
|
| 37 |
+
# Initial convolution block
|
| 38 |
model0 = [ nn.ReflectionPad2d(3),
|
| 39 |
nn.Conv2d(input_nc, 64, 7),
|
| 40 |
norm_layer(64),
|
| 41 |
nn.ReLU(inplace=True) ]
|
| 42 |
self.model0 = nn.Sequential(*model0)
|
| 43 |
|
| 44 |
+
# Downsampling
|
| 45 |
model1 = []
|
| 46 |
in_features = 64
|
| 47 |
out_features = in_features*2
|
|
|
|
| 53 |
out_features = in_features*2
|
| 54 |
self.model1 = nn.Sequential(*model1)
|
| 55 |
|
| 56 |
+
# Residual blocks
|
| 57 |
model2 = []
|
| 58 |
for _ in range(n_residual_blocks):
|
| 59 |
model2 += [ResidualBlock(in_features)]
|
| 60 |
self.model2 = nn.Sequential(*model2)
|
| 61 |
|
| 62 |
+
# Upsampling
|
| 63 |
model3 = []
|
| 64 |
out_features = in_features//2
|
| 65 |
for _ in range(2):
|
|
|
|
| 70 |
out_features = in_features//2
|
| 71 |
self.model3 = nn.Sequential(*model3)
|
| 72 |
|
| 73 |
+
# Output layer
|
| 74 |
model4 = [ nn.ReflectionPad2d(3),
|
| 75 |
+
nn.Conv2d(64, output_nc, 7)]
|
| 76 |
if sigmoid:
|
| 77 |
model4 += [nn.Sigmoid()]
|
| 78 |
+
|
| 79 |
self.model4 = nn.Sequential(*model4)
|
| 80 |
|
| 81 |
+
def forward(self, x):
|
| 82 |
out = self.model0(x)
|
| 83 |
out = self.model1(out)
|
| 84 |
out = self.model2(out)
|
| 85 |
out = self.model3(out)
|
| 86 |
out = self.model4(out)
|
|
|
|
| 87 |
return out
|
| 88 |
|
| 89 |
+
# Initialize models
|
| 90 |
+
def load_models():
|
| 91 |
+
model1 = Generator(3, 1, 3).to(device)
|
| 92 |
+
model2 = Generator(3, 1, 3).to(device)
|
| 93 |
+
|
| 94 |
+
# Download models from HuggingFace Hub
|
| 95 |
+
model1_path = hf_hub_download(repo_id="your-hf-repo/line-drawing", filename="model.pth")
|
| 96 |
+
model2_path = hf_hub_download(repo_id="your-hf-repo/line-drawing", filename="model2.pth")
|
| 97 |
+
|
| 98 |
+
model1.load_state_dict(torch.load(model1_path, map_location=device))
|
| 99 |
+
model2.load_state_dict(torch.load(model2_path, map_location=device))
|
| 100 |
+
|
| 101 |
+
model1.eval()
|
| 102 |
+
model2.eval()
|
| 103 |
+
return model1, model2
|
| 104 |
+
|
| 105 |
+
model1, model2 = load_models()
|
| 106 |
+
|
| 107 |
+
def apply_style_transfer(img, strength=1.0):
|
| 108 |
+
"""Apply artistic style transfer effect"""
|
| 109 |
+
img_array = np.array(img)
|
| 110 |
+
processed = F.interpolate(
|
| 111 |
+
torch.from_numpy(img_array).float().unsqueeze(0),
|
| 112 |
+
size=(256, 256),
|
| 113 |
+
mode='bilinear',
|
| 114 |
+
align_corners=False
|
| 115 |
+
)
|
| 116 |
+
return processed * strength
|
| 117 |
+
|
| 118 |
+
def enhance_lines(img, contrast=1.0, brightness=1.0):
|
| 119 |
+
"""Enhance line drawing with contrast and brightness adjustments"""
|
| 120 |
+
enhanced = np.array(img)
|
| 121 |
+
enhanced = enhanced * contrast
|
| 122 |
+
enhanced = np.clip(enhanced + brightness, 0, 1)
|
| 123 |
+
return Image.fromarray((enhanced * 255).astype(np.uint8))
|
| 124 |
+
|
| 125 |
+
def predict(input_img, version, line_thickness=1.0, contrast=1.0, brightness=1.0, enable_enhancement=False):
|
| 126 |
+
try:
|
| 127 |
+
# Open and process input image
|
| 128 |
+
original_img = Image.open(input_img)
|
| 129 |
+
original_size = original_img.size
|
| 130 |
+
|
| 131 |
+
# Transform pipeline
|
| 132 |
+
transform = transforms.Compose([
|
| 133 |
+
transforms.Resize(256, Image.BICUBIC),
|
| 134 |
+
transforms.ToTensor(),
|
| 135 |
+
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))
|
| 136 |
+
])
|
| 137 |
+
|
| 138 |
+
input_tensor = transform(original_img).unsqueeze(0).to(device)
|
| 139 |
+
|
| 140 |
+
# Process through selected model
|
| 141 |
+
with torch.no_grad():
|
| 142 |
+
if version == 'Simple Lines':
|
| 143 |
+
output = model2(input_tensor)
|
| 144 |
+
else:
|
| 145 |
+
output = model1(input_tensor)
|
| 146 |
+
|
| 147 |
+
# Apply line thickness adjustment
|
| 148 |
+
output = output * line_thickness
|
| 149 |
+
|
| 150 |
+
# Convert to image
|
| 151 |
+
output_img = transforms.ToPILImage()(output.squeeze().cpu().clamp(0, 1))
|
| 152 |
+
|
| 153 |
+
# Apply enhancements if enabled
|
| 154 |
+
if enable_enhancement:
|
| 155 |
+
output_img = enhance_lines(output_img, contrast, brightness)
|
| 156 |
+
|
| 157 |
+
# Resize to original
|
| 158 |
+
output_img = output_img.resize(original_size, Image.BICUBIC)
|
| 159 |
+
|
| 160 |
+
return output_img
|
| 161 |
+
|
| 162 |
+
except Exception as e:
|
| 163 |
+
raise gr.Error(f"Error processing image: {str(e)}")
|
| 164 |
+
|
| 165 |
+
# Custom CSS for better UI
|
| 166 |
+
custom_css = """
|
| 167 |
+
.gradio-container {
|
| 168 |
+
font-family: 'Helvetica Neue', Arial, sans-serif;
|
| 169 |
+
}
|
| 170 |
+
.gr-button {
|
| 171 |
+
border-radius: 8px;
|
| 172 |
+
background: linear-gradient(45deg, #3498db, #2980b9);
|
| 173 |
+
border: none;
|
| 174 |
+
color: white;
|
| 175 |
+
}
|
| 176 |
+
.gr-button:hover {
|
| 177 |
+
background: linear-gradient(45deg, #2980b9, #3498db);
|
| 178 |
+
transform: translateY(-2px);
|
| 179 |
+
transition: all 0.3s ease;
|
| 180 |
+
}
|
| 181 |
+
.gr-input {
|
| 182 |
+
border-radius: 8px;
|
| 183 |
+
border: 2px solid #3498db;
|
| 184 |
+
}
|
| 185 |
+
"""
|
| 186 |
+
|
| 187 |
+
# Create Gradio interface with enhanced UI
|
| 188 |
+
with gr.Blocks(css=custom_css) as iface:
|
| 189 |
+
gr.Markdown("# 🎨 Advanced Line Drawing Generator")
|
| 190 |
+
gr.Markdown("Transform your images into beautiful line drawings with advanced controls")
|
| 191 |
+
|
| 192 |
+
with gr.Row():
|
| 193 |
+
with gr.Column():
|
| 194 |
+
input_image = gr.Image(type="filepath", label="Upload Image")
|
| 195 |
+
version = gr.Radio(
|
| 196 |
+
choices=['Complex Lines', 'Simple Lines'],
|
| 197 |
+
value='Simple Lines',
|
| 198 |
+
label="Drawing Style"
|
| 199 |
+
)
|
| 200 |
+
|
| 201 |
+
with gr.Accordion("Advanced Settings", open=False):
|
| 202 |
+
line_thickness = gr.Slider(
|
| 203 |
+
minimum=0.1,
|
| 204 |
+
maximum=2.0,
|
| 205 |
+
value=1.0,
|
| 206 |
+
step=0.1,
|
| 207 |
+
label="Line Thickness"
|
| 208 |
+
)
|
| 209 |
+
enable_enhancement = gr.Checkbox(
|
| 210 |
+
label="Enable Enhancement",
|
| 211 |
+
value=False
|
| 212 |
+
)
|
| 213 |
+
with gr.Group(visible=False) as enhancement_controls:
|
| 214 |
+
contrast = gr.Slider(
|
| 215 |
+
minimum=0.5,
|
| 216 |
+
maximum=2.0,
|
| 217 |
+
value=1.0,
|
| 218 |
+
step=0.1,
|
| 219 |
+
label="Contrast"
|
| 220 |
+
)
|
| 221 |
+
brightness = gr.Slider(
|
| 222 |
+
minimum=0.5,
|
| 223 |
+
maximum=1.5,
|
| 224 |
+
value=1.0,
|
| 225 |
+
step=0.1,
|
| 226 |
+
label="Brightness"
|
| 227 |
+
)
|
| 228 |
+
|
| 229 |
+
enable_enhancement.change(
|
| 230 |
+
fn=lambda x: gr.Group(visible=x),
|
| 231 |
+
inputs=[enable_enhancement],
|
| 232 |
+
outputs=[enhancement_controls]
|
| 233 |
+
)
|
| 234 |
+
|
| 235 |
+
with gr.Column():
|
| 236 |
+
output_image = gr.Image(type="pil", label="Generated Line Drawing")
|
| 237 |
+
|
| 238 |
+
with gr.Row():
|
| 239 |
+
generate_btn = gr.Button("Generate Drawing", variant="primary")
|
| 240 |
+
clear_btn = gr.Button("Clear", variant="secondary")
|
| 241 |
+
|
| 242 |
+
# Load example images
|
| 243 |
+
example_images = []
|
| 244 |
+
for file in os.listdir('.'):
|
| 245 |
+
if file.lower().endswith(('.png', '.jpg', '.jpeg')):
|
| 246 |
+
example_images.append(file)
|
| 247 |
+
|
| 248 |
+
if example_images:
|
| 249 |
+
gr.Examples(
|
| 250 |
+
examples=[[img, "Simple Lines"] for img in example_images],
|
| 251 |
+
inputs=[input_image, version],
|
| 252 |
+
outputs=output_image,
|
| 253 |
+
fn=predict,
|
| 254 |
+
cache_examples=True
|
| 255 |
+
)
|
| 256 |
+
|
| 257 |
+
# Set up event handlers
|
| 258 |
+
generate_btn.click(
|
| 259 |
+
fn=predict,
|
| 260 |
+
inputs=[
|
| 261 |
+
input_image,
|
| 262 |
+
version,
|
| 263 |
+
line_thickness,
|
| 264 |
+
contrast,
|
| 265 |
+
brightness,
|
| 266 |
+
enable_enhancement
|
| 267 |
+
],
|
| 268 |
+
outputs=output_image
|
| 269 |
+
)
|
| 270 |
+
|
| 271 |
+
clear_btn.click(
|
| 272 |
+
fn=lambda: (None, "Simple Lines", 1.0, 1.0, 1.0, False),
|
| 273 |
+
inputs=[],
|
| 274 |
+
outputs=[
|
| 275 |
+
input_image,
|
| 276 |
+
version,
|
| 277 |
+
line_thickness,
|
| 278 |
+
contrast,
|
| 279 |
+
brightness,
|
| 280 |
+
enable_enhancement
|
| 281 |
+
]
|
| 282 |
+
)
|
| 283 |
|
| 284 |
+
# Launch the interface
|
| 285 |
iface.launch()
|