app.py

import gradio as gr
import torch
from torchvision import transforms
from PIL import Image
import torch.nn as nn

# ==========================
# Konfigurace modelu
# ==========================
model_path = "models/frUn_OR_13k0050_512.pth"  # Model bude nahrán do složky "models" na Hugging Face
target_width = 1500

# ==========================
# Definice U-Net modelu
# ==========================
class UNet(nn.Module):
    def __init__(self, n_channels=3, n_classes=3, bilinear=True):
        super(UNet, self).__init__()
        self.n_channels = n_channels
        self.n_classes = n_classes
        self.bilinear = bilinear

        self.inc = double_conv(n_channels, 64)
        self.down1 = nn.Sequential(
            nn.MaxPool2d(2),
            double_conv(64, 128)
        )
        self.down2 = nn.Sequential(
            nn.MaxPool2d(2),
            double_conv(128, 256)
        )
        self.down3 = nn.Sequential(
            nn.MaxPool2d(2),
            double_conv(256, 512)
        )
        self.down4 = nn.Sequential(
            nn.MaxPool2d(2),
            double_conv(512, 512)
        )
        self.up1 = Up(512 + 512, 256, bilinear)
        self.up2 = Up(256 + 256, 128, bilinear)
        self.up3 = Up(128 + 128, 64, bilinear)
        self.up4 = Up(64 + 64, 64, bilinear)
        self.outc = nn.Conv2d(64, n_classes, kernel_size=1)
        self.final_activation = nn.Sigmoid()

    def forward(self, x):
        x1 = self.inc(x)
        x2 = self.down1(x1)
        x3 = self.down2(x2)
        x4 = self.down3(x3)
        x5 = self.down4(x4)

        x = self.up1(x5, x4)
        x = self.up2(x, x3)
        x = self.up3(x, x2)
        x = self.up4(x, x1)

        x = self.outc(x)
        x = self.final_activation(x)
        return x

def double_conv(in_channels, out_channels):
    return nn.Sequential(
        nn.Conv2d(in_channels, out_channels, kernel_size=3, padding=1),
        nn.BatchNorm2d(out_channels),
        nn.ReLU(inplace=True),
        nn.Conv2d(out_channels, out_channels, kernel_size=3, padding=1),
        nn.BatchNorm2d(out_channels),
        nn.ReLU(inplace=True),
    )

class Up(nn.Module):
    def __init__(self, in_channels, out_channels, bilinear=True):
        super(Up, self).__init__()
        if bilinear:
            self.up = nn.Upsample(scale_factor=2, mode='bilinear', align_corners=True)
            self.conv = double_conv(in_channels, out_channels)
        else:
            self.up = nn.ConvTranspose2d(in_channels // 2, in_channels // 2, kernel_size=2, stride=2)
            self.conv = double_conv(in_channels, out_channels)

    def forward(self, x1, x2):
        x1 = self.up(x1)
        diffY = x2.size()[2] - x1.size()[2]
        diffX = x2.size()[3] - x1.size()[3]
        x1 = nn.functional.pad(x1, [diffX // 2, diffX - diffX // 2,
                                    diffY // 2, diffY - diffY // 2])
        x = torch.cat([x2, x1], dim=1)
        return self.conv(x)

# ==========================
# Načtení modelu
# ==========================
device = torch.device("cpu")  # Vždy používáme CPU
model = UNet(n_channels=3, n_classes=3, bilinear=True).to(device)
model.load_state_dict(torch.load(model_path, map_location=device), strict=False)
model.eval()

# ==========================
# Pomocná funkce pro změnu velikosti s zachováním poměru stran
# ==========================
def resize_keeping_aspect_ratio(image, target_width):
    aspect_ratio = image.height / image.width
    target_height = int(target_width * aspect_ratio)
    return image.resize((target_width, target_height), Image.LANCZOS)

# ==========================
# Funkce pro převod obrázku modelem
# ==========================
def generate_sketch(input_image):
    # Změna velikosti na požadovanou šířku
    resized_image = resize_keeping_aspect_ratio(input_image, target_width=target_width)
    
    # Převod obrázku na tensor
    transform = transforms.ToTensor()
    image_tensor = transform(resized_image).unsqueeze(0).to(device)
    
    # Provedení převodu modelem
    with torch.no_grad():
        output = model(image_tensor).squeeze(0).cpu()
    
    # Převod výsledného tensoru zpět na PIL Image
    output_image = transforms.ToPILImage()(output)
    return output_image

# ==========================
# Rozhraní Gradio
# ==========================
interface = gr.Interface(
    fn=generate_sketch,
    inputs=gr.Image(type="pil", label="Vstupní obrázek"),
    outputs=gr.Image(type="pil", format="jpeg"),
    title="U-Net Image2draw",
    description=""
)

interface.launch()