app.py

from PIL import Image, ImageDraw, ImageFont
import numpy as np
import cv2
import tensorflow as tf
import gradio as gr
import io

def load_model(model_path):
    model = tf.keras.models.load_model(model_path)
    model.compile(optimizer=tf.keras.optimizers.Adam(), loss=tf.keras.losses.BinaryCrossentropy(), metrics=['accuracy'])
    return model

def get_model_summary(model):
    stream = io.StringIO()
    model.summary(print_fn=lambda x: stream.write(x + "\n"))
    summary_str = stream.getvalue()
    stream.close()
    return summary_str

def get_input_shape(model):
    input_shape = model.input_shape[1:]  # Skip the batch dimension
    return input_shape

def preprocess_image(image, input_shape):
    img = np.array(image)
    num_channels = input_shape[-1]

    if num_channels == 1:  # Model expects grayscale
        if len(img.shape) == 2:  # Image is already grayscale
            img = np.expand_dims(img, axis=-1)
        elif img.shape[2] == 3:  # Convert RGB to grayscale
            img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
            img = np.expand_dims(img, axis=-1)
    elif num_channels == 3:  # Model expects RGB
        if len(img.shape) == 2:  # Convert grayscale to RGB
            img = cv2.cvtColor(img, cv2.COLOR_GRAY2RGB)
        elif img.shape[2] == 1:  # Convert single channel to RGB
            img = cv2.cvtColor(img, cv2.COLOR_GRAY2RGB)
    
    img_resized = cv2.resize(img, (input_shape[0], input_shape[1]))
    img_normalized = img_resized / 255.0
    img_batch = np.expand_dims(img_normalized, axis=0)
    
    return img_batch

def diagnose_image(image, model, input_shape):
    img_batch = preprocess_image(image, input_shape)
    prediction = model.predict(img_batch)
    glaucoma_probability = prediction[0][0]
    result_text = f"Probability of glaucoma: {glaucoma_probability:.2%}"
    
    img_display = np.array(image)
    if len(img_display.shape) == 2 or img_display.shape[2] == 1:  # Convert to RGB for display
        img_display = cv2.cvtColor(img_display, cv2.COLOR_GRAY2RGB)
    image_pil = Image.fromarray(img_display)
    draw = ImageDraw.Draw(image_pil)
    font = ImageFont.load_default()
    
    text = f"{glaucoma_probability:.2%}"
    text_size = draw.textsize(text, font=font)
    
    rect_width = 200
    rect_height = 100
    rect_x = (image_pil.width - rect_width) // 2
    rect_y = (image_pil.height - rect_height) // 2
    
    draw.rectangle([rect_x, rect_y, rect_x + rect_width, rect_y + rect_height], outline="red", width=3)
    
    text_x = rect_x + (rect_width - text_size[0]) // 2
    text_y = rect_y + (rect_height - text_size[1]) // 2
    
    draw.text((text_x, text_y), text, fill="red", font=font)
    
    return image_pil, result_text

def main():
    with gr.Blocks() as demo:
        gr.Markdown("# Glaucoma Detection App")
        gr.Markdown("Upload a fundus eye image to detect the probability of glaucoma.")
        
        with gr.Row():
            model_file = gr.File(label="Upload Model (.h5 or .keras)")
            load_model_btn = gr.Button("Load Model")
            model_info = gr.Markdown()
        
        image = gr.Image(type="pil", label="Upload Image")
        submit_btn = gr.Button("Diagnose")
        result = gr.Textbox(label="Diagnosis Result")
        
        def load_and_display_model_info(file):
            model = load_model(file.name)
            model_summary = get_model_summary(model)
            input_shape = get_input_shape(model)
            return model, model_summary, input_shape
        
        model = gr.State(None)
        input_shape = gr.State(None)
        
        def diagnose_and_display(image, model, input_shape):
            diagnosis_image, result_text = diagnose_image(image, model, input_shape)
            return diagnosis_image, result_text
        
        load_model_btn.click(fn=load_and_display_model_info, inputs=model_file, outputs=[model, model_info, input_shape])
        submit_btn.click(fn=diagnose_and_display, inputs=[image, model, input_shape], outputs=[image, result])
        
        gr.Markdown("### Glaucoma Analyzer V.1.0.0 by Thariq Arian")
        
    demo.launch()

if __name__ == "__main__":
    main()