app.py

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

# CPU 전용 설정
torch.set_num_threads(4)  # CPU 스레드 수 제한
torch.set_grad_enabled(False)  # 추론 모드만 사용

norm_layer = nn.InstanceNorm2d

class ResidualBlock(nn.Module):
    def __init__(self, in_features):
        super(ResidualBlock, self).__init__()
        conv_block = [
            nn.ReflectionPad2d(1),
            nn.Conv2d(in_features, in_features, 3),
            norm_layer(in_features),
            nn.ReLU(inplace=True),
            nn.ReflectionPad2d(1),
            nn.Conv2d(in_features, in_features, 3),
            norm_layer(in_features)
        ]
        self.conv_block = nn.Sequential(*conv_block)

    def forward(self, x):
        return x + self.conv_block(x)

class Generator(nn.Module):
    def __init__(self, input_nc, output_nc, n_residual_blocks=9, sigmoid=True):
        super(Generator, self).__init__()
        
        # Initial convolution block
        model0 = [
            nn.ReflectionPad2d(3),
            nn.Conv2d(input_nc, 64, 7),
            norm_layer(64),
            nn.ReLU(inplace=True)
        ]
        self.model0 = nn.Sequential(*model0)

        # Downsampling
        model1 = []
        in_features = 64
        out_features = in_features*2
        for _ in range(2):
            model1 += [
                nn.Conv2d(in_features, out_features, 3, stride=2, padding=1),
                norm_layer(out_features),
                nn.ReLU(inplace=True)
            ]
            in_features = out_features
            out_features = in_features*2
        self.model1 = nn.Sequential(*model1)

        # Residual blocks
        model2 = []
        for _ in range(n_residual_blocks):
            model2 += [ResidualBlock(in_features)]
        self.model2 = nn.Sequential(*model2)

        # Upsampling
        model3 = []
        out_features = in_features//2
        for _ in range(2):
            model3 += [
                nn.ConvTranspose2d(in_features, out_features, 3, stride=2, padding=1, output_padding=1),
                norm_layer(out_features),
                nn.ReLU(inplace=True)
            ]
            in_features = out_features
            out_features = in_features//2
        self.model3 = nn.Sequential(*model3)

        # Output layer
        model4 = [
            nn.ReflectionPad2d(3),
            nn.Conv2d(64, output_nc, 7)
        ]
        if sigmoid:
            model4 += [nn.Sigmoid()]
        self.model4 = nn.Sequential(*model4)

    def forward(self, x):
        out = self.model0(x)
        out = self.model1(out)
        out = self.model2(out)
        out = self.model3(out)
        out = self.model4(out)
        return out

# CPU 전용 모델 로드
def load_models():
    try:
        print("Initializing models in CPU mode...")
        model1 = Generator(3, 1, 3)
        model2 = Generator(3, 1, 3)
        
        # Load models in CPU mode
        model1.load_state_dict(torch.load('model.pth', map_location='cpu'))
        model2.load_state_dict(torch.load('model2.pth', map_location='cpu'))
        
        model1.eval()
        model2.eval()
        
        print("Models loaded successfully")
        return model1, model2
    except Exception as e:
        print(f"Error loading models: {str(e)}")
        raise gr.Error("Failed to initialize models. Please check model files.")

try:
    print("Starting model initialization...")
    model1, model2 = load_models()
    print("Model initialization completed")
except Exception as e:
    print(f"Critical error: {str(e)}")
    raise gr.Error("Failed to start the application")

def process_image(input_img, version, line_thickness=1.0):
    try:
        # 이미지 로드 및 전처리
        original_img = Image.open(input_img)
        original_size = original_img.size
        
        transform = transforms.Compose([
            transforms.Resize(256, Image.BICUBIC),
            transforms.ToTensor(),
            transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))
        ])
        
        input_tensor = transform(original_img).unsqueeze(0)
        
        # 모델 처리
        with torch.no_grad():
            if version == 'Simple Lines':
                output = model2(input_tensor)
            else:
                output = model1(input_tensor)
            
            output = output * line_thickness
        
        # 결과 이미지 생성
        output_img = transforms.ToPILImage()(output.squeeze().clamp(0, 1))
        output_img = output_img.resize(original_size, Image.BICUBIC)
        
        return output_img
        
    except Exception as e:
        raise gr.Error(f"이미지 처리 에러: {str(e)}")

# Simple UI
with gr.Blocks() as iface:
    gr.Markdown("# ✨ Magic Drawings")
    gr.Markdown("Transform your photos into magical line art with AI")
    
    with gr.Row():
        with gr.Column():
            input_image = gr.Image(type="filepath", label="Upload Image")
            version = gr.Radio(
                choices=['Complex Lines', 'Simple Lines'],
                value='Simple Lines',
                label="Art Style"
            )
            line_thickness = gr.Slider(
                minimum=0.1,
                maximum=2.0,
                value=1.0,
                step=0.1,
                label="Line Thickness"
            )
            
        with gr.Column():
            output_image = gr.Image(type="pil", label="Generated Art")
    
    generate_btn = gr.Button("Generate Magic", variant="primary")
    
    # Event handlers
    generate_btn.click(
        fn=process_image,
        inputs=[input_image, version, line_thickness],
        outputs=output_image
    )

# 실행
iface.launch(
    server_name="0.0.0.0",
    server_port=7860,
    share=False
)