File size: 4,880 Bytes
a60a7da
146cb65
 
 
 
 
 
 
a60a7da
146cb65
 
 
 
 
 
a60a7da
146cb65
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
a60a7da
146cb65
a60a7da
146cb65
 
 
 
 
a60a7da
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
146cb65
 
 
a60a7da
 
146cb65
 
a60a7da
 
 
 
 
146cb65
 
a60a7da
 
 
 
 
 
146cb65
 
 
 
 
 
 
a60a7da
 
 
146cb65
a60a7da
 
 
 
 
 
146cb65
 
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

import torch
import torch.nn as nn
import torch.optim as optim
from torchvision import models, transforms
from torchvision.models import VGG19_Weights
from PIL import Image
import gradio as gr
import time

# ✅ Use GPU if available
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
print("Using device:", device)

# --- Image Utilities ---
def load_image(img, max_size=384):
    transform = transforms.Compose([
        transforms.Resize(max_size),
        transforms.ToTensor(),
        transforms.Normalize(mean=[0.485, 0.456, 0.406],
                             std=[0.229, 0.224, 0.225])
    ])
    image = img.convert('RGB')
    image = transform(image).unsqueeze(0)
    return image.to(device)

def tensor_to_image(tensor):
    unnormalize = transforms.Normalize(
        mean=[-0.485 / 0.229, -0.456 / 0.224, -0.406 / 0.225],
        std=[1 / 0.229, 1 / 0.224, 1 / 0.225]
    )
    image = tensor.clone().detach().squeeze(0)
    image = unnormalize(image)
    image = torch.clamp(image, 0, 1)
    return transforms.ToPILImage()(image)

# --- Style Transfer Utilities ---
def gram_matrix(tensor):
    b, c, h, w = tensor.size()
    features = tensor.view(b * c, h * w)
    return torch.mm(features, features.t())

class StyleTransferNet(nn.Module):
    def __init__(self, style_img, content_img):
        super().__init__()
        weights = VGG19_Weights.DEFAULT
        self.vgg = models.vgg19(weights=weights).features.to(device).eval()
        self.style_img = style_img
        self.content_img = content_img
        self.content_layers = ['conv_4']
        self.style_layers = ['conv_1', 'conv_3', 'conv_5', 'conv_9']

    def get_features(self, x):
        features = {}
        i = 0
        for layer in self.vgg.children():
            x = layer(x)
            if isinstance(layer, nn.Conv2d):
                i += 1
                name = f'conv_{i}'
                if name in self.content_layers + self.style_layers:
                    features[name] = x
        return features

    def forward(self, input_img, steps=100, style_weight=1e6, content_weight=1e5):
        input_img = input_img.clone().requires_grad_(True)
        optimizer = optim.Adam([input_img], lr=0.02)

        style_features = self.get_features(self.style_img)
        content_features = self.get_features(self.content_img)
        style_grams = {k: gram_matrix(v) for k, v in style_features.items()}

        for step in range(steps):
            optimizer.zero_grad()
            target_features = self.get_features(input_img)
            style_loss = 0
            content_loss = 0

            for layer in self.style_layers:
                target_feature = target_features[layer]
                target_gram = gram_matrix(target_feature)
                style_gram = style_grams[layer]
                style_loss += torch.mean((target_gram - style_gram)**2)

            for layer in self.content_layers:
                target_feature = target_features[layer]
                content_feature = content_features[layer]
                content_loss += torch.mean((target_feature - content_feature)**2)

            total_loss = style_weight * style_loss + content_weight * content_loss
            total_loss.backward()
            optimizer.step()

        return input_img

# --- Gradio App ---
def style_transfer_app(content_img, style_img, content_weight_ui, style_weight_ui, steps):
    start_time = time.time()
    content = load_image(content_img)
    style = load_image(style_img)

    # Map intuitive UI weights (1-10) to actual values
    content_weight = content_weight_ui * 1e5
    style_weight = style_weight_ui * 1e6

    model = StyleTransferNet(style, content)
    output = model(content, steps=int(steps), content_weight=content_weight, style_weight=style_weight)
    stylized = tensor_to_image(output)
    elapsed = round(time.time() - start_time)

    # Estimated time display
    estimate_note = f"🕒 Estimated processing time: {elapsed} seconds for {steps} steps."
    return stylized, estimate_note

# --- Launch Interface ---
gr.Interface(
    fn=style_transfer_app,
    inputs=[
        gr.Image(type="pil", label="🖼️ Content Image"),
        gr.Image(type="pil", label="🎨 Style Image"),
        gr.Slider(1, 10, value=1, step=1, label="Content Weight (1 = weak structure, 10 = strong)"),
        gr.Slider(1, 10, value=6, step=1, label="Style Weight (1 = subtle, 10 = strong style)"),
        gr.Slider(50, 300, value=100, step=50, label="Steps (speed vs quality)")
    ],
    outputs=[
        gr.Image(type="pil", label="🧠 Stylized Output"),
        gr.Textbox(label="⏱️ Time Info")
    ],
    title="🎨 Fast AI Neural Style Transfer",
    description="Upload content and style images, then tune how much structure vs style you want. Powered by PyTorch + VGG19.",
    allow_flagging="never"
).launch(share=True)