app.py

import gradio as gr
import torch
import torch.nn as nn
import torch.nn.functional as F
from torchvision import transforms, models
from PIL import Image
import os
import time

# Define the transform for the input image
transform = transforms.Compose([
    transforms.Resize((150, 150)),
    transforms.ToTensor(),
    transforms.Normalize((0.5,), (0.5,))
])

# Load the trained ResNet50 model
class FineTunedResNet(nn.Module):
    def __init__(self, num_classes=4):
        super(FineTunedResNet, self).__init__()
        self.resnet = models.resnet50(weights=models.ResNet50_Weights.DEFAULT)  # Load pre-trained ResNet50

        # Replace the fully connected layer with more layers and batch normalization
        self.resnet.fc = nn.Sequential(
            nn.Linear(self.resnet.fc.in_features, 1024),  # First additional layer
            nn.BatchNorm1d(1024),
            nn.ReLU(),
            nn.Dropout(0.5),
            nn.Linear(1024, 512),  # Second additional layer
            nn.BatchNorm1d(512),
            nn.ReLU(),
            nn.Dropout(0.5),
            nn.Linear(512, 256),  # Third additional layer
            nn.BatchNorm1d(256),
            nn.ReLU(),
            nn.Dropout(0.5),
            nn.Linear(256, num_classes)  # Output layer
        )

    def forward(self, x):
        return self.resnet(x)

model = FineTunedResNet(num_classes=4)
model_path = 'models/final_fine_tuned_resnet50.pth'

if not os.path.exists(model_path):
    raise FileNotFoundError(f"The model file '{model_path}' does not exist. Please check the path.")

model.load_state_dict(torch.load(model_path, map_location=torch.device('cpu')))
model.eval()

# Define a function to make predictions
def predict(image):
    start_time = time.time()  # Start the timer
    image = transform(image).unsqueeze(0)  # Transform and add batch dimension
    
    with torch.no_grad():
        output = model(image)
        probabilities = F.softmax(output, dim=1)[0]
        top_prob, top_class = torch.topk(probabilities, 3)
        classes = ['🦠 COVID', '🫁 Normal', '🦠 Pneumonia', '🦠 TB']  # Adjust based on the classes in your model
        
    end_time = time.time()  # End the timer
    prediction_time = end_time - start_time  # Calculate the prediction time

    # Format the result string
    result = f"Top Predictions:\n"
    for i in range(top_prob.size(0)):
        result += f"{classes[top_class[i]]}: Score {top_prob[i].item()}\n"
    result += f"Prediction Time: {prediction_time:.2f} seconds"
    
    return result

# Example images with labels
examples = [
    ['examples/Pneumonia/02009view1_frontal.jpg', '🦠 Pneumonia'],
    ['examples/Pneumonia/02055view1_frontal.jpg', '🦠 Pneumonia'],
    ['examples/Pneumonia/03152view1_frontal.jpg', '🦠 Pneumonia'],
    ['examples/COVID/11547_2020_1200_Fig3_HTML-a.png', '🦠 COVID'],
    ['examples/COVID/11547_2020_1200_Fig3_HTML-b.png', '🦠 COVID'],
    ['examples/COVID/11547_2020_1203_Fig1_HTML-b.png', '🦠 COVID'],
    ['examples/Normal/06bc1cfe-23a0-43a4-a01b-dfa10314bbb0.jpg', '🫁 Normal'],
    ['examples/Normal/08ae6c0b-d044-4de2-a410-b3cf8dc65868.jpg', '🫁 Normal'],
    ['examples/Normal/IM-0178-0001.jpeg', '🫁 Normal']
]

# Load visualization images
visualization_images = [
    "pictures/1.png",
    "pictures/2.png",
    "pictures/3.png",
    "pictures/4.png",
    "pictures/5.png"
]

# Function to display visualization images
def display_visualizations():
    return [Image.open(image) for image in visualization_images]

# Custom CSS to enhance appearance (injected via HTML)
custom_css = """
    <style>
    body {
        font-family: 'Arial', sans-serif;
        background-color: #f5f5f5;
    }
    .gradio-container {
        background-color: #ffffff;
        border: 1px solid #e6e6e6;
        box-shadow: 0 4px 8px rgba(0, 0, 0, 0.1);
        border-radius: 10px;
        padding: 20px;
    }
    .gradio-title {
        color: #333333;
        font-weight: bold;
        font-size: 24px;
        margin-bottom: 10px;
    }
    .gradio-description {
        color: #666666;
        font-size: 16px;
        margin-bottom: 20px;
    }
    .gradio-image {
        border-radius: 10px;
    }
    .gradio-button {
        background-color: #007bff;
        color: #ffffff;
        border: none;
        padding: 10px 20px;
        border-radius: 5px;
        cursor: pointer;
    }
    .gradio-button:hover {
        background-color: #0056b3;
    }
    .gradio-label {
        color: #007bff;
        font-weight: bold;
    }
    </style>
"""

# Create Gradio interfaces
prediction_interface = gr.Interface(
    fn=predict,
    inputs=gr.Image(type="pil", label="Upload Chest X-ray Image"),
    outputs=gr.Label(label="Predicted Disease"),
    examples=examples,
    title="Lung Disease Detection XVI",
    description=f"""
        Upload a chest X-ray image to detect lung diseases such as 🦠 COVID-19, 🦠 Pneumonia, 🫁 Normal, or 🦠 TB. 
        Use the example images to see how the model works.
        {custom_css}
    """
)

visualization_interface = gr.Interface(
    fn=display_visualizations,
    inputs=None,
    outputs=[gr.Image(type="pil", label=f"Visualization {i+1}") for i in range(len(visualization_images))],
    title="Model Performance Visualizations",
    description=f"""
        Here are some visualizations that depict the performance of the model during training and testing.
        {custom_css}
    """
)

# Combine interfaces into a tabbed interface
app = gr.TabbedInterface(
    interface_list=[prediction_interface, visualization_interface],
    tab_names=["Predict", "Model Performance"]
)

# Launch the interface
app.launch(share=True)