src/streamlit_app.py

import streamlit as st
from streamlit_webrtc import webrtc_streamer, WebRtcMode, RTCConfiguration
import av
import cv2
import numpy as np
import torch
import torch.nn.functional as F
import json
from huggingface_hub import hf_hub_download
from collections import deque
import plotly.graph_objects as go
from PIL import Image

# Page config
st.set_page_config(
    page_title="MindSense AI | Emotion Recognition",
    page_icon="🧠",
    layout="wide"
)

# Custom CSS
st.markdown("""
<style>
    @import url('https://fonts.googleapis.com/css2?family=Inter:wght@400;600;700&display=swap');
    
    * { font-family: 'Inter', sans-serif; }
    
    .main {
        background: linear-gradient(135deg, #667eea 0%, #764ba2 100%);
    }
    
    .title-gradient {
        background: linear-gradient(90deg, #667eea 0%, #764ba2 50%, #f093fb 100%);
        -webkit-background-clip: text;
        -webkit-text-fill-color: transparent;
        font-size: 3rem;
        font-weight: 800;
        text-align: center;
        margin-bottom: 10px;
    }
    
    .subtitle {
        text-align: center;
        color: rgba(255, 255, 255, 0.9);
        font-size: 1.1rem;
        margin-bottom: 30px;
    }
    
    .metric-card {
        background: rgba(255, 255, 255, 0.1);
        backdrop-filter: blur(20px);
        border: 1px solid rgba(255, 255, 255, 0.2);
        border-radius: 15px;
        padding: 20px;
        margin: 10px 0;
    }
    
    div[data-testid="stMetricValue"] {
        font-size: 1.8rem;
        font-weight: 700;
    }
</style>
""", unsafe_allow_html=True)

# ============================================================================
# Load Model from HuggingFace Hub
# ============================================================================

@st.cache_resource
def load_model():
    """Download and load model from HF Hub"""
    repo_id = "Arko007/mindsense-emotion-model"
    
    with st.spinner("🧠 Loading AI model..."):
        try:
            model_path = hf_hub_download(repo_id=repo_id, filename="mindsense_emotion_model.pt")
            config_path = hf_hub_download(repo_id=repo_id, filename="model_config.json")
            
            with open(config_path, 'r') as f:
                config = json.load(f)
            
            model = torch.jit.load(model_path, map_location='cpu')
            model.eval()
            
            return model, config
            
        except Exception as e:
            st.error(f"❌ Error loading model: {e}")
            return None, None

model, config = load_model()

if model is None:
    st.error("Failed to load model. Please check the repository.")
    st.stop()

st.success(f"✅ Model loaded! Accuracy: {config.get('best_val_acc', 0):.2f}%")

# ============================================================================
# Emotion Analyzer
# ============================================================================

class EmotionAnalyzer:
    def __init__(self, model, config):
        self.model = model
        self.config = config
        self.emotions = config['classes']
        self.mean = np.array(config['mean']).reshape(3, 1, 1)
        self.std = np.array(config['std']).reshape(3, 1, 1)
        self.face_cascade = cv2.CascadeClassifier(
            cv2.data.haarcascades + 'haarcascade_frontalface_default.xml'
        )
    
    @torch.no_grad()
    def analyze_frame(self, frame):
        """Analyze frame for emotions"""
        try:
            gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
            faces = self.face_cascade.detectMultiScale(gray, 1.3, 5)
            
            if len(faces) == 0:
                return self._default_result()
            
            x, y, w, h = max(faces, key=lambda f: f[2] * f[3])
            face_roi = frame[y:y+h, x:x+w]
            
            # Preprocess
            face_rgb = cv2.cvtColor(face_roi, cv2.COLOR_BGR2RGB)
            face_resized = cv2.resize(face_rgb, (384, 384))
            
            img_tensor = torch.from_numpy(face_resized).float().permute(2, 0, 1) / 255.0
            img_tensor = (img_tensor - torch.from_numpy(self.mean).float()) / torch.from_numpy(self.std).float()
            img_tensor = img_tensor.unsqueeze(0)
            
            # Inference
            emotion_logits, stress_pred, valence_pred = self.model(img_tensor)
            
            emotion_probs = F.softmax(emotion_logits, dim=1)[0].numpy()
            emotion_idx = np.argmax(emotion_probs)
            
            return {
                'dominant_emotion': self.emotions[emotion_idx],
                'confidence': float(emotion_probs[emotion_idx]),
                'all_emotions': {e: float(p) for e, p in zip(self.emotions, emotion_probs)},
                'stress_score': float(stress_pred.item()),
                'valence': float(valence_pred.item()),
                'face_location': (x, y, w, h)
            }
            
        except Exception as e:
            return self._default_result()
    
    def _default_result(self):
        return {
            'dominant_emotion': 'neutral',
            'confidence': 0.0,
            'all_emotions': {e: 0.0 for e in self.emotions},
            'stress_score': 0.0,
            'valence': 0.0,
            'face_location': None
        }

# Initialize analyzer
if 'analyzer' not in st.session_state:
    st.session_state.analyzer = EmotionAnalyzer(model, config)
if 'emotion_history' not in st.session_state:
    st.session_state.emotion_history = deque(maxlen=100)
if 'stress_scores' not in st.session_state:
    st.session_state.stress_scores = deque(maxlen=100)

# ============================================================================
# UI
# ============================================================================

st.markdown('<h1 class="title-gradient">🧠 MindSense AI</h1>', unsafe_allow_html=True)
st.markdown('<p class="subtitle">Real-Time Emotion Recognition & Mental Health Assessment</p>', unsafe_allow_html=True)

# Sidebar
with st.sidebar:
    st.markdown("### ⚙️ Settings")
    confidence_threshold = st.slider("Confidence Threshold", 0.0, 1.0, 0.5, 0.05)
    show_all_emotions = st.checkbox("Show All Emotions", value=True)
    
    st.markdown("---")
    st.markdown("### 📊 Model Info")
    st.info(f"""
    **Architecture:** Custom EfficientNet-CNN
    
    **Parameters:** {config.get('total_params', 0) / 1e6:.2f}M
    
    **Accuracy:** {config.get('best_val_acc', 0):.2f}%
    
    **Trained on:** FER2013 (28k images)
    """)

# Main content
tab1, tab2 = st.tabs(["🎥 Live Webcam", "📤 Upload Image"])

with tab1:
    col1, col2 = st.columns([2, 1])
    
    with col1:
        st.markdown("### Live Analysis")
        
        rtc_config = RTCConfiguration(
            {"iceServers": [{"urls": ["stun:stun.l.google.com:19302"]}]}
        )
        
        class VideoProcessor:
            def __init__(self):
                self.frame_count = 0
                
            def recv(self, frame):
                img = frame.to_ndarray(format="bgr24")
                self.frame_count += 1
                
                if self.frame_count % 3 == 0:
                    result = st.session_state.analyzer.analyze_frame(img)
                    
                    if result['face_location']:
                        x, y, w, h = result['face_location']
                        emotion = result['dominant_emotion']
                        confidence = result['confidence']
                        
                        color_map = {
                            'happy': (0, 255, 0), 'sad': (255, 0, 0),
                            'angry': (0, 0, 255), 'fear': (128, 0, 128),
                            'surprise': (255, 255, 0), 'neutral': (128, 128, 128),
                            'disgust': (0, 128, 128)
                        }
                        color = color_map.get(emotion, (255, 255, 255))
                        
                        cv2.rectangle(img, (x, y), (x+w, y+h), color, 2)
                        label = f"{emotion.upper()} ({confidence:.0%})"
                        cv2.putText(img, label, (x, y-10), cv2.FONT_HERSHEY_SIMPLEX, 0.7, color, 2)
                        
                        if confidence > confidence_threshold:
                            st.session_state.emotion_history.append(emotion)
                            st.session_state.stress_scores.append(result['stress_score'])
                
                return av.VideoFrame.from_ndarray(img, format="bgr24")
        
        webrtc_ctx = webrtc_streamer(
            key="emotion-detection",
            mode=WebRtcMode.SENDRECV,
            rtc_configuration=rtc_config,
            video_processor_factory=VideoProcessor,
            media_stream_constraints={"video": True, "audio": False},
            async_processing=True
        )
    
    with col2:
        st.markdown("### 📊 Live Metrics")
        
        if len(st.session_state.emotion_history) > 0:
            current_emotion = st.session_state.emotion_history[-1]
            avg_stress = np.mean(list(st.session_state.stress_scores)[-10:])
            
            emotion_emoji = {
                'happy': '😊', 'sad': '😢', 'angry': '😠',
                'fear': '😨', 'surprise': '😮', 'neutral': '😐',
                'disgust': '🤢'
            }
            
            st.markdown(f"## {emotion_emoji.get(current_emotion, '😐')} {current_emotion.title()}")
            st.metric("Stress Level", f"{avg_stress:.1%}")
            st.progress(avg_stress)
            
            if show_all_emotions:
                st.markdown("#### All Emotions")
                result = st.session_state.analyzer.analyze_frame(np.zeros((100, 100, 3), dtype=np.uint8))
                for emotion, prob in sorted(result['all_emotions'].items(), key=lambda x: x[1], reverse=True):
                    st.text(f"{emotion.title()}: {prob:.1%}")
        else:
            st.info("👋 Start webcam to begin")

with tab2:
    st.markdown("### Upload an Image")
    uploaded_file = st.file_uploader("Choose an image...", type=['jpg', 'jpeg', 'png'])
    
    if uploaded_file:
        image = Image.open(uploaded_file)
        image_np = np.array(image)
        
        col1, col2 = st.columns(2)
        
        with col1:
            st.image(image, caption="Uploaded Image", use_column_width=True)
        
        with col2:
            result = st.session_state.analyzer.analyze_frame(cv2.cvtColor(image_np, cv2.COLOR_RGB2BGR))
            
            st.markdown("### 🎭 Analysis Results")
            st.markdown(f"**Emotion:** {result['dominant_emotion'].title()}")
            st.markdown(f"**Confidence:** {result['confidence']:.1%}")
            st.markdown(f"**Stress:** {result['stress_score']:.1%}")
            st.markdown(f"**Valence:** {result['valence']:.2f}")
            
            if show_all_emotions:
                st.markdown("#### Emotion Distribution")
                for emotion, prob in sorted(result['all_emotions'].items(), key=lambda x: x[1], reverse=True):
                    st.progress(prob)
                    st.caption(f"{emotion.title()}: {prob:.1%}")

# Visualizations
if len(st.session_state.emotion_history) > 10:
    st.markdown("---")
    st.markdown("### 📈 Analysis Dashboard")
    
    col1, col2 = st.columns(2)
    
    with col1:
        from collections import Counter
        emotion_counts = Counter(st.session_state.emotion_history)
        
        fig = go.Figure(data=[go.Pie(
            labels=list(emotion_counts.keys()),
            values=list(emotion_counts.values()),
            hole=0.4
        )])
        fig.update_layout(title="Emotion Distribution", height=300)
        st.plotly_chart(fig, use_container_width=True)
    
    with col2:
        fig = go.Figure()
        fig.add_trace(go.Scatter(
            y=list(st.session_state.stress_scores),
            mode='lines',
            fill='tozeroy',
            line=dict(color='#667eea', width=2)
        ))
        fig.update_layout(title="Stress Timeline", height=300, yaxis_range=[0, 1])
        st.plotly_chart(fig, use_container_width=True)

# Footer
st.markdown("---")
st.markdown("""
<div style='text-align:center; color:rgba(255,255,255,0.7);'>
    <p>🧠 MindSense AI | Built with PyTorch & Streamlit</p>
    <p>⚠️ <strong>Disclaimer:</strong> Research tool only. Not for medical diagnosis.</p>
</div>
""", unsafe_allow_html=True)