Upload app.py

app.py

@@ -0,0 +1,366 @@
+import gradio as gr
+import tensorflow as tf
+import numpy as np
+import cv2
+from PIL import Image
+import os
+
+# ==============================
+# REGISTER CUSTOM LAYERS
+# ==============================
+import tensorflow as tf
+from tensorflow.keras.layers import Layer
+from tensorflow.keras.utils import register_keras_serializable
+
+@register_keras_serializable(package="Custom")
+class ChannelMeanPooling(Layer):
+    def call(self, inputs):
+        return tf.reduce_mean(inputs, axis=3, keepdims=True)
+
+@register_keras_serializable(package="Custom")
+class ChannelMaxPooling(Layer):
+    def call(self, inputs):
+        return tf.reduce_max(inputs, axis=3, keepdims=True)
+
+
+# ==============================
+# CONFIGURATION
+# ==============================
+IMG_SIZE = 224
+CLASS_NAMES = ["cataracts", "diabetic retinopathy", "glaucoma", "normal"]
+
+MODEL_CONFIG = {
+	"VGG16": {
+		"path": os.path.join("models", "SpaAtt_vgg16_model.keras"),
+		"preprocess": tf.keras.applications.vgg16.preprocess_input,
+	},
+	"Inception-v3": {
+		"path": os.path.join("models", "SpaAtt_inceptionv3_model.keras"),
+		"preprocess": tf.keras.applications.inception_v3.preprocess_input,
+	},
+	"ResNet50": {
+		"path": os.path.join("models", "SpaAtt_resnet50_model.keras"),
+		"preprocess": tf.keras.applications.resnet50.preprocess_input,
+	},
+	"DenseNet121": {
+		"path": os.path.join("models", "SpaAtt_densenet121_model.keras"),
+		"preprocess": tf.keras.applications.densenet.preprocess_input,
+	},
+	"EfficientNet-B0": {
+		"path": os.path.join("models", "SpaAtt_efficientnetb0_model.keras"),
+		"preprocess": tf.keras.applications.efficientnet.preprocess_input,
+	},
+}
+
+
+# ==============================
+# LOAD MODELS (ONCE)
+# ==============================
+custom_objects = {
+    "ChannelMeanPooling": ChannelMeanPooling,
+    "ChannelMaxPooling": ChannelMaxPooling,
+}
+
+MODELS = {}
+
+for name, cfg in MODEL_CONFIG.items():
+	if not os.path.exists(cfg["path"]):
+		raise FileNotFoundError(f"Model not found: {cfg['path']}")
+	MODELS[name] = tf.keras.models.load_model(
+        cfg["path"],
+        custom_objects = custom_objects,
+        safe_mode = False
+    )
+	print(f"Loaded model: {name}")
+
+	
+# ==============================
+# IMAGE PROCESSING UTILITIES
+# ==============================
+def extract_retinal_fov(img_rgb):
+	"""
+	Extract circular retinal field of view using brightness mask.
+	"""
+	gray = cv2.cvtColor(img_rgb, cv2.COLOR_RGB2GRAY)
+	_, thresh = cv2.threshold(gray, 15, 255, cv2.THRESH_BINARY)
+	
+	contours, _ = cv2.findContours(thresh, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
+	if not contours:
+		return img_rgb
+		
+	c = max(contours, key = cv2.contourArea)
+	x, y, w, h = cv2.boundingRect(c)
+	return img_rgb[y:y+h, x:x+w]
+	
+def clahe_l_channel(img_rgb, clip = 2.0):
+	lab = cv2.cvtColor(img_rgb, cv2.COLOR_RGB2LAB)
+	l, a, b = cv2.split(lab)
+	clahe = cv2.createCLAHE(clipLimit = clip, tileGridSize = (8,8))
+	l = clahe.apply(l)
+	lab = cv2.merge((l, a, b))
+	return cv2.cvtColor(lab, cv2.COLOR_LAB2RGB)
+	
+def enhance_vessels(img_rgb, ksize = 15):
+	gray = cv2.cvtColor(img_rgb, cv2.COLOR_RGB2GRAY)
+	kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (ksize, ksize))
+	top_hat = cv2.morphologyEx(gray, cv2.MORPH_TOPHAT, kernel)
+	black_hat = cv2.morphologyEx(gray, cv2.MORPH_BLACKHAT, kernel)
+	enhanced = cv2.add(gray, top_hat)
+	enhanced = cv2.subtract(enhanced, black_hat)
+	enhanced = cv2.normalize(enhanced, None, 0, 255, cv2.NORM_MINMAX)
+	return cv2.cvtColor(enhanced, cv2.COLOR_GRAY2RGB)
+	
+def enhance_optic_disc(img_rgb, ksize = 30):
+	gray = cv2.cvtColor(img_rgb, cv2.COLOR_RGB2GRAY)
+	kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (ksize, ksize))
+	closing = cv2.morphologyEx(gray, cv2.MORPH_CLOSE, kernel)
+	disc = cv2.subtract(gray, closing)
+	disc = cv2.normalize(disc, None, 0, 255, cv2.NORM_MINMAX)
+	return cv2.cvtColor(disc, cv2.COLOR_GRAY2RGB)
+	
+def full_enhancement_pipeline(img_rgb):
+	img = extract_retinal_fov(img_rgb)
+	img = cv2.resize(img, (IMG_SIZE, IMG_SIZE))
+	img = clahe_l_channel(img)
+	vessels = enhance_vessels(img)
+	disc = enhance_optic_disc(img)
+	
+	#Blend enhancements
+	img = cv2.addWeighted(img, 0.85, vessels, 0.15, 0)
+	img = cv2.addWeighted(img, 0.85, disc, 0.15, 0)
+	return img
+	
+	
+# ==============================
+# ENSEMBLE PREDICTION
+# ==============================
+def ensemble_predict(input_image):
+	if input_image is None:
+		return None, "❌ No image uploaded."
+
+	#Convert to RGB
+	img_rgb = np.array(input_image.convert("RGB"))
+	img = full_enhancement_pipeline(img_rgb)
+
+	probs = []
+	
+	for model_name, model in MODELS.items():
+		preprocess = MODEL_CONFIG[model_name]["preprocess"]
+		
+		img_input = np.expand_dims(img, axis = 0)
+		img_input = preprocess(img_input.astype(np.float32))
+		
+		pred = model.predict(img_input, verbose = 0)[0]
+		probs.append(pred)
+		
+	#Soft Voting (Average Probabilities)
+	probs = np.array(probs)
+	mean_probs = probs.mean(axis = 0)
+	
+	result = {CLASS_NAMES[i]: float(mean_probs[i]) for i in range(len(CLASS_NAMES))}
+	predicted_class_name = CLASS_NAMES[int(np.argmax(mean_probs))]
+	
+	return result, f"✅ Prediction: **{predicted_class_name.upper()}**"
+	
+	
+# ==============================
+# GRADIO UI
+# ==============================
+js = """
+function createGradioAnimation() {
+    const run = () => {
+        var container = document.createElement('div');
+        container.id = 'gradio-animation';
+        container.style.fontSize = '2em';
+        container.style.fontWeight = 'bold';
+        container.style.textAlign = 'center';
+        container.style.marginBottom = '20px';
+    
+        var text = 'A-EYE: An Intelligent Eye Disease Classifier';
+        
+        for (var i = 0; i < text.length; i++) {
+            setTimeout(function(){
+                var letter = document.createElement('span');
+                letter.style.opacity = '0';
+                letter.style.transition = 'opacity 0.5s';
+                letter.innerText = text[i];
+    
+                container.appendChild(letter);
+    
+                setTimeout(() {
+                    letter.style.opacity = '1';
+                }, 50);
+            }, i * 100);
+        }
+        const gradioContainer = document.querySelector('.gradio-container');
+        if (gradioContainer){
+            gradioContainer.insertBefore(container, gradioContainer.firstChild);
+        }
+    };
+    window.addEventListener("load", function(){
+        createGradioAnimation();
+    });
+}
+"""
+
+css = """
+#banner-img img {
+    width: 100% !important;
+    height: auto !important;
+    max-height: 280px;
+    object-fit: contain;
+}
+
+#input-img img {
+    width: 300px !important;
+    height: 300px !important;
+    object-fit: contain;
+}
+
+#output-label {
+    height: 300px !important;
+    display: flex;
+    flex-direction: column;
+    justify-content: center; /*vertical alignment*/
+}
+
+#input-box img {
+    width: 100%;
+    height: 100%;
+    object-fit: contain;  /* preserves aspect ratio */
+}
+
+#output-box > div {
+    height: 100%;
+    display: flex;
+    flex-direction: column;
+    justify-content: center;
+}
+
+/* Hide all image action buttons (download, share, etc.) */
+button[aria-label="Download"],
+button[aria-label="Share"],
+button[aria-label="Open in new tab"] {
+    display: none !important;
+}
+
+/* Also hide top-right image toolbar if present */
+.gradio-container .absolute.top-0.right-0 {
+    display: none !important;
+}
+"""
+
+
+#Background
+background_img = Image.open('background.png').resize((3000, 700))
+
+with gr.Blocks() as demo:
+    gr.HTML("""
+    <link href="https://fonts.googleapis.com/css2?family=Poppins:wght@600&display=swap" rel="stylesheet">
+    
+    <style>
+    @keyframes fadeIn{
+        from { opacity: 0; transform: translateY(-10px); }
+        to { opacity: 1; transform: translateY(0);}
+    }
+    .title {
+        font-family: 'Poppins', sans-serif;
+        text-align: center;
+        font-size: 36px;
+        color: #2c3e50;
+        animation: fadeIn 1.5s ease-in-out;
+    }
+    </style>
+    
+    <div class="title">
+        A-EYE: An Intelligent Eye Disease Classifier
+    </div>
+    """)
+    
+    with gr.Row():
+        gr.Image(
+            background_img, 
+            interactive = False, 
+            elem_id = "banner-img",
+            #show_download_button = False, 
+            #show_share_button = False,
+        )
+        
+    with gr.Row():
+        gr.Markdown(
+            """
+            <div style="text-align: justify;">
+            A-EYE is an intelligent eye disease classifier designed for accurate eye disease classification using deep learning.
+            With its enhanced model, A-EYE analyzes fundus images to detect Cataracts, Diabetic Retinopathy, Glaucoma, or Normal conditions.
+            Users can easily upload a fundus image, and the system will process it to provide detailed classification results. 
+            The results display probability percentages for each condition, ensuring transparency and confidence in the diagnosis.
+            The condition with the highest probability is assigned to the image, offering a reliable and efficient tool for eye health assessment.
+            A-EYE has the potential to assist ophthalmologists in providing first screening of suspected eye diseases.
+            </div>
+            """
+        )
+        # gr.HTML("""
+        #     <div style="
+        #         #max-width: 1000px;
+        #         width: 100%
+        #         margin: 10px auto;
+        #         padding: 10px 15px;
+        #         background: #f8fafc;
+        #         border-radius: 12px;
+        #         box-shadow: 0 4px 12px rgba(0,0,0,0.05);
+        #         font-family: 'Inter', sans-serif;
+        #         line-height: 1.6;
+        #         color: #2c3e50;
+        #     ">
+        #         <h3 style="text-align: center; margin-bottom: 15px;">
+        #                 🔍 About A-EYE
+        #         </h3>
+                
+        #         <p style="text-align: justify;">
+        #             <b>A-EYE</b> is an intelligent eye disease classifier designed for accurate eye disease classification using 
+        #             <b>five attention-enhanced deep learning models</b> combined via a 
+        #             <span style="color:#2563eb; font-weight:600;">soft-voting ensemble</span>.
+        #         </p>
+        #         <hr style="border: none; border-top: 1px solid #e0e0e0; margin: 15px 0;">
+        #         <p style="text-align: justify;">
+        #             It analyzes fundus images to detect:
+        #             <b>🟡 Cataracts</b>, <b>🔴 Diabetic Retinopathy</b>, <b>🔵 Glaucoma</b>, or <b>🟢 Normal</b> conditions.
+        #         </p>
+        #         <hr style="border: none; border-top: 1px solid #e0e0e0; margin: 15px 0;">
+        #         <p style="text-align: justify;">
+        #             User can easily upload a fundus image, and A-EYE will generate classification results with <b>probability scores</b> for each condition -  ensuring transparency and confidence.
+        #         </p>
+                
+        #         <p style="text-align: justify;">
+        #             The highest probability determines the final prediction, offering a reliable tool for <b>early screening and decision support</b>.
+        #         </p>
+        #         <hr style="border: none; border-top: 1px solid #e0e0e0; margin: 15px 0;">
+        #         <p style="text-align: justify;">
+        #            💡 <i>A-EYE is designed to assist ophthalmologists in providing first screening of eye diseases.</i>
+        #         </p>
+        #     </div>
+        # """
+        # )
+        
+    with gr.Row():
+        with gr.Column():	
+            img_input = gr.Image(
+                type = "pil", 
+                label = "Upload Fundus Image",
+                elem_id = "input-img"
+            )
+            predict_button = gr.Button("🔍 Predict")
+        with gr.Column():
+            output_label = gr.Label(
+                num_top_classes = 4,
+                elem_id = "output-label"
+            )
+            output_text = gr.Markdown()
+        
+        predict_button.click(
+            fn = ensemble_predict,
+            inputs = img_input,
+            outputs = [output_label, output_text]
+        )
+
+demo.launch(debug = True, share = True, theme=gr.themes.Soft(), css=css, js = js)