import gradio as gr import cv2 import numpy as np import tensorflow as tf import face_recognition # ========================================================================= # 1. Load Trained CNN Model & Haar Cascade # ========================================================================= print("Loading Model...") try: smile_model = tf.keras.models.load_model('smile_model.h5') print("✅ Model loaded successfully!") except Exception as e: print(f"❌ Error loading model: {e}. Please ensure 'smile_model.h5' is uploaded.") smile_model = None face_cascade = cv2.CascadeClassifier(cv2.data.haarcascades + 'haarcascade_frontalface_default.xml') # ========================================================================= # 2. Helper Functions (Exactly from your Colab) # ========================================================================= def mouth_aspect_ratio(landmarks): left = np.array(landmarks['top_lip'][0]) right = np.array(landmarks['top_lip'][6]) upper = np.array(landmarks['top_lip'][3]) lower = np.array(landmarks['bottom_lip'][3]) # calculate MAR return np.linalg.norm(right-left) / (np.linalg.norm(lower-upper) + 1e-6) # ========================================================================= # 3. Main Processing Function # ========================================================================= def detect_smile(image, method, cnn_threshold, mar_threshold): if image is None: return None, "No image provided." # Gradio provides image as RGB numpy array frame = image.copy() gray = cv2.cvtColor(frame, cv2.COLOR_RGB2GRAY) faces = face_cascade.detectMultiScale(gray, 1.1, 5) smile_count = 0 total_faces = len(faces) if method == "CNN Model (from .h5)": if smile_model is None: return frame, "Error: CNN Model not loaded." for (x, y, w, h) in faces: face_roi = frame[y:y+h, x:x+w] face_resized = cv2.resize(face_roi, (224, 224)) face_arr = np.expand_dims(face_resized / 255.0, axis=0).astype(np.float32) prob = smile_model.predict(face_arr, verbose=0)[0][0] label = 'Smile' if prob > cnn_threshold else 'No Smile' color = (0, 255, 0) if label == 'Smile' else (255, 0, 0) # Gradio Uses RGB, so Red is (255,0,0) if label == 'Smile': smile_count += 1 cv2.rectangle(frame, (x, y), (x+w, y+h), color, 2) cv2.putText(frame, f"{label} ({prob:.2f})", (x, y-10), cv2.FONT_HERSHEY_SIMPLEX, 0.7, color, 2) elif method == "Mouth Aspect Ratio (MAR)": # Using face_recognition library as in Cell 13 landmarks_list = face_recognition.face_landmarks(frame) for (x, y, w, h) in faces: ratio = 0.0 is_smile = False for lm in landmarks_list: # Check if this landmark belongs to the current face bounding box nose = np.array(lm['nose_tip'][0]) if x <= nose[0] <= x+w and y <= nose[1] <= y+h: ratio = mouth_aspect_ratio(lm) is_smile = ratio > mar_threshold break label = 'Smile' if is_smile else 'No Smile' color = (0, 255, 0) if is_smile else (255, 0, 0) if is_smile: smile_count += 1 cv2.rectangle(frame, (x, y), (x+w, y+h), color, 2) cv2.putText(frame, f"{label} (r={ratio:.1f})", (x, y-10), cv2.FONT_HERSHEY_SIMPLEX, 0.7, color, 2) status = f"Total Faces Detected: {total_faces} | Smiles Detected: {smile_count}" return frame, status # ========================================================================= # 4. Gradio Interface Layout # ========================================================================= with gr.Blocks(theme=gr.themes.Soft()) as demo: gr.Markdown("# 😃 Smile Detector Pro") gr.Markdown("Upload an image or use your webcam to detect smiles. Choose between your Custom Trained CNN or the Mouth Aspect Ratio (MAR) method.") with gr.Row(): with gr.Column(): input_image = gr.Image(sources=["upload", "webcam"], type="numpy", label="Input Image / Live Camera") method_radio = gr.Radio(["CNN Model (from .h5)", "Mouth Aspect Ratio (MAR)"], value="CNN Model (from .h5)", label="Detection Method") cnn_thresh = gr.Slider(0.0, 1.0, value=0.85, step=0.01, label="CNN Threshold", info="Default: 0.85 (Cell 11)") mar_thresh = gr.Slider(1.0, 5.0, value=3.5, step=0.1, label="MAR Threshold", info="Default: 3.5 (Cell 13)") submit_btn = gr.Button("🔍 Detect Smile", variant="primary") with gr.Column(): output_image = gr.Image(type="numpy", label="Result") output_text = gr.Textbox(label="Status") submit_btn.click( fn=detect_smile, inputs=[input_image, method_radio, cnn_thresh, mar_thresh], outputs=[output_image, output_text] ) # Run the app demo.launch()