app.py

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

# ==========================================
# 1. MODEL ARCHITECTURE (Same as training)
# ==========================================
class ESPCN(nn.Module):
    def __init__(self, upscale_factor):
        super(ESPCN, self).__init__()
        self.conv1 = nn.Conv2d(3, 64, kernel_size=5, padding=2)
        self.conv2 = nn.Conv2d(64, 64, kernel_size=3, padding=1)
        self.conv3 = nn.Conv2d(64, 32, kernel_size=3, padding=1)
        self.conv4 = nn.Conv2d(32, 3 * (upscale_factor ** 2), kernel_size=3, padding=1)
        self.pixel_shuffle = nn.PixelShuffle(upscale_factor)
        self.relu = nn.ReLU()

    def forward(self, x):
        x = self.relu(self.conv1(x))
        x = self.relu(self.conv2(x))
        x = self.relu(self.conv3(x))
        x = self.pixel_shuffle(self.conv4(x))
        return x

# ==========================================
# 2. LOAD THE TRAINED MODEL (On CPU)
# ==========================================
# Device ko CPU set kar rahe hain taaki HF Spaces ke free tier me bina GPU ke chale
device = torch.device("cpu")
model = ESPCN(upscale_factor=4)

# Yahan hum .pth file load kar rahe hain
try:
    model.load_state_dict(torch.load("universal_sr_model.pth", map_location=device))
    model.eval()
    print("Model loaded successfully!")
except Exception as e:
    print(f"Error loading model: {e}")

# ==========================================
# 3. INFERENCE FUNCTION FOR GRADIO
# ==========================================
def enhance_image(img):
    if img is None:
        return None
        
    # Image ko RGB me ensure karna
    img = img.convert('RGB')
    
    # Image ko tensor me convert karna aur batch dimension add karna
    input_tensor = transforms.ToTensor()(img).unsqueeze(0).to(device)
    
    # Model se pass karna (no gradients needed for fast inference)
    with torch.no_grad():
        output_tensor = model(input_tensor)
        
    # Output tensor ko wapas image me convert karna
    output_tensor = output_tensor.squeeze().clamp(0, 1)
    output_img = transforms.ToPILImage()(output_tensor)
    
    return output_img

# ==========================================
# 4. GRADIO INTERFACE SETUP
# ==========================================
# UI design setup
iface = gr.Interface(
    fn=enhance_image,
    inputs=gr.Image(type="pil", label="Upload Unclear/Low-Res Image"),
    outputs=gr.Image(type="pil", label="4x Enhanced High-Res Image"),
    title="🌟 AI Super Resolution (4x Upscaler)",
    description="Apni unclear ya chhoti image upload karein aur hamara custom ESPCN model use 4x bada aur sharp kar dega. Yeh model scratch se train kiya gaya hai!",
    # 'allow_flagging' line hata di gayi hai taaki latest Gradio me crash na ho
    flagging_mode="never" 
)

# App launch karna
if __name__ == "__main__":
    iface.launch()