Initial commit

app.py

@@ -0,0 +1,59 @@
+import gradio as gr
+import cv2
+import numpy as np
+import pickle
+from util import get_face_landmarks
+
+# Emotion labels
+emotions = ['HAPPY', 'SAD', 'SURPRISED']
+
+# Load the model once (not inside the function for speed)
+with open('model.pkl', 'rb') as f:
+    model = pickle.load(f)
+
+def predict_emotion(image):
+    # Convert PIL image to OpenCV format
+    img = cv2.cvtColor(np.array(image), cv2.COLOR_RGB2BGR)
+
+    # Extract landmarks
+    face_landmarks = get_face_landmarks(img, draw=False, static_image_mode=True)
+
+    if face_landmarks is None:
+        return {"No face detected": 1.0}, image
+
+    # Predict emotion and probabilities
+    output = model.predict([face_landmarks])
+    predicted_emotion = emotions[int(output[0])]
+    
+    if hasattr(model, "predict_proba"):
+        probs = model.predict_proba([face_landmarks])[0]
+        confidence_dict = {emotions[i]: float(probs[i]) for i in range(len(emotions))}
+    else:
+        confidence_dict = {predicted_emotion: 1.0}
+
+    # Annotate image
+    cv2.putText(img, predicted_emotion, (10, img.shape[0]-10),
+                cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 255, 0), 2)
+
+    # Convert back to RGB for Gradio
+    return confidence_dict, cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
+
+# Example images (ensure these exist in /examples folder)
+examples = [["examples/happy.png"], ["examples/sad.png"], ["examples/surprised.png"]]
+
+# Gradio Interface
+demo = gr.Interface(
+    fn=predict_emotion,
+    inputs=gr.Image(type="pil", label="Upload Image or Use Webcam", sources=["upload", "webcam"]),
+    outputs=[
+        gr.Label(num_top_classes=3, label="Predicted Emotion & Confidence"),
+        gr.Image(type="numpy", label="Annotated Image")
+    ],
+    title="Emotion Detector",
+    description="Upload an image or use webcam to detect emotions (HAPPY, SAD, SURPRISED).",
+    examples=examples,
+    theme="default"
+)
+
+if __name__ == "__main__":
+    demo.launch()

model.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:766b2b73840a165978ed0258c4d77a7ff8b34129d5fac5b30d4b8c3503f8b354
+size 18684691

requirements.txt

@@ -0,0 +1,5 @@
+tensorflow==2.15.0
+mediapipe==0.10.21
+numpy==1.26.4
+opencv-python-headless==4.8.1.78
+scikit-learn==1.3.2

util.py

@@ -0,0 +1,46 @@
+import cv2
+import mediapipe as mp
+
+def get_face_landmarks(image, draw=False, static_image_mode=True):
+
+    # Read the input image
+    image_input_rgb = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
+
+    face_mesh = mp.solutions.face_mesh.FaceMesh(static_image_mode=static_image_mode,
+                                                max_num_faces=1,
+                                                min_detection_confidence=0.5)
+    image_rows, image_cols, _ = image.shape
+    results = face_mesh.process(image_input_rgb)
+
+    image_landmarks = []
+
+    if results.multi_face_landmarks:
+
+        if draw:
+
+            mp_drawing = mp.solutions.drawing_utils
+            mp_drawing_styles = mp.solutions.drawing_styles
+            drawing_spec = mp_drawing.DrawingSpec(thickness=2, circle_radius=1)
+
+            mp_drawing.draw_landmarks(
+                image=image,
+                landmark_list=results.multi_face_landmarks[0],
+                connections=mp.solutions.face_mesh.FACEMESH_CONTOURS,
+                landmark_drawing_spec=drawing_spec,
+                connection_drawing_spec=drawing_spec)
+
+        ls_single_face = results.multi_face_landmarks[0].landmark
+        xs_ = []
+        ys_ = []
+        zs_ = []
+        for idx in ls_single_face:
+            xs_.append(idx.x)
+            ys_.append(idx.y)
+            zs_.append(idx.z)
+        for j in range(len(xs_)):
+            image_landmarks.append(xs_[j] - min(xs_))
+            image_landmarks.append(ys_[j] - min(ys_))
+            image_landmarks.append(zs_[j] - min(zs_))
+
+    return image_landmarks
+