Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -15,7 +15,6 @@ class EmotionCNN(nn.Module):
|
|
| 15 |
def __init__(self):
|
| 16 |
super(EmotionCNN, self).__init__()
|
| 17 |
self.conv_layers = nn.Sequential(
|
| 18 |
-
# Premier bloc
|
| 19 |
nn.Conv2d(1, 32, 3, padding=1),
|
| 20 |
nn.BatchNorm2d(32),
|
| 21 |
nn.ReLU(),
|
|
@@ -24,8 +23,6 @@ class EmotionCNN(nn.Module):
|
|
| 24 |
nn.ReLU(),
|
| 25 |
nn.MaxPool2d(2),
|
| 26 |
nn.Dropout2d(0.25),
|
| 27 |
-
|
| 28 |
-
# Deuxième bloc
|
| 29 |
nn.Conv2d(32, 64, 3, padding=1),
|
| 30 |
nn.BatchNorm2d(64),
|
| 31 |
nn.ReLU(),
|
|
@@ -34,8 +31,6 @@ class EmotionCNN(nn.Module):
|
|
| 34 |
nn.ReLU(),
|
| 35 |
nn.MaxPool2d(2),
|
| 36 |
nn.Dropout2d(0.25),
|
| 37 |
-
|
| 38 |
-
# Troisième bloc
|
| 39 |
nn.Conv2d(64, 128, 3, padding=1),
|
| 40 |
nn.BatchNorm2d(128),
|
| 41 |
nn.ReLU(),
|
|
@@ -45,7 +40,6 @@ class EmotionCNN(nn.Module):
|
|
| 45 |
nn.MaxPool2d(2),
|
| 46 |
nn.Dropout2d(0.25)
|
| 47 |
)
|
| 48 |
-
|
| 49 |
self.fc_layers = nn.Sequential(
|
| 50 |
nn.Linear(128 * 6 * 6, 512),
|
| 51 |
nn.ReLU(),
|
|
@@ -62,7 +56,7 @@ class EmotionCNN(nn.Module):
|
|
| 62 |
x = self.fc_layers(x)
|
| 63 |
return x
|
| 64 |
|
| 65 |
-
# Dictionnaire des émotions
|
| 66 |
emotion_dict = {
|
| 67 |
0: {"name": "Colère", "message": "Respirez profondément et prenez un moment pour vous calmer."},
|
| 68 |
1: {"name": "Mépris", "message": "Essayez de voir les choses d'un autre point de vue."},
|
|
@@ -129,8 +123,12 @@ st.title("🎭 Détecteur d'Émotions en Temps Réel")
|
|
| 129 |
def load_model():
|
| 130 |
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
| 131 |
model = EmotionCNN().to(device)
|
| 132 |
-
|
| 133 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 134 |
return model, device
|
| 135 |
|
| 136 |
# Chargement du modèle
|
|
@@ -146,15 +144,31 @@ transform = transforms.Compose([
|
|
| 146 |
|
| 147 |
# Chargement du classificateur Haar pour la détection de visage
|
| 148 |
face_cascade = cv2.CascadeClassifier(cv2.data.haarcascades + 'haarcascade_frontalface_default.xml')
|
|
|
|
|
|
|
|
|
|
| 149 |
|
| 150 |
def detect_faces(frame):
|
| 151 |
gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
|
| 152 |
faces = face_cascade.detectMultiScale(gray, 1.1, 4)
|
| 153 |
return faces
|
| 154 |
|
| 155 |
-
# Configuration RTC
|
| 156 |
RTC_CONFIGURATION = RTCConfiguration({
|
| 157 |
-
"iceServers": [
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 158 |
})
|
| 159 |
|
| 160 |
# Classe pour traiter les frames vidéo
|
|
@@ -169,55 +183,61 @@ class VideoProcessor(VideoProcessorBase):
|
|
| 169 |
self.message_placeholder = st.session_state.get('message_placeholder')
|
| 170 |
|
| 171 |
def recv(self, frame):
|
| 172 |
-
|
| 173 |
-
|
| 174 |
-
|
| 175 |
-
for (x, y, w, h) in faces:
|
| 176 |
-
cv2.rectangle(img, (x, y), (x+w, y+h), (0, 255, 0), 2)
|
| 177 |
-
face_img = img[y:y+h, x:x+w]
|
| 178 |
-
pil_img = Image.fromarray(cv2.cvtColor(face_img, cv2.COLOR_BGR2RGB))
|
| 179 |
-
img_tensor = self.transform(pil_img).unsqueeze(0).to(self.device)
|
| 180 |
-
with torch.no_grad():
|
| 181 |
-
output = self.model(img_tensor)
|
| 182 |
-
_, predicted = torch.max(output, 1)
|
| 183 |
-
emotion_idx = predicted.item()
|
| 184 |
-
emotion_name = self.emotion_dict[emotion_idx]["name"]
|
| 185 |
-
cv2.putText(img, emotion_name, (x, y-10),
|
| 186 |
-
cv2.FONT_HERSHEY_SIMPLEX, 0.9, (0, 255, 0), 2)
|
| 187 |
-
|
| 188 |
-
# Mettre à jour les placeholders (utiliser st.session_state pour partager)
|
| 189 |
-
if self.emotion_placeholder:
|
| 190 |
-
self.emotion_placeholder.markdown(f"""
|
| 191 |
-
<div class="emotion-box">
|
| 192 |
-
<div class="emotion-title">{emotion_name}</div>
|
| 193 |
-
</div>
|
| 194 |
-
""", unsafe_allow_html=True)
|
| 195 |
|
| 196 |
-
|
| 197 |
-
|
| 198 |
-
|
| 199 |
-
|
| 200 |
-
|
| 201 |
-
|
| 202 |
-
|
| 203 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 204 |
|
| 205 |
-
# Configuration de
|
| 206 |
col1, col2 = st.columns([2, 1])
|
| 207 |
|
| 208 |
with col1:
|
| 209 |
st.markdown("### 📹 Flux Vidéo")
|
| 210 |
-
|
| 211 |
-
|
| 212 |
-
|
| 213 |
-
|
| 214 |
-
|
| 215 |
-
|
| 216 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 217 |
|
| 218 |
with col2:
|
| 219 |
st.markdown("### 😊 Émotion Détectée")
|
| 220 |
-
# Utiliser session_state pour partager les placeholders avec VideoProcessor
|
| 221 |
if 'emotion_placeholder' not in st.session_state:
|
| 222 |
st.session_state.emotion_placeholder = st.empty()
|
| 223 |
if 'message_placeholder' not in st.session_state:
|
|
@@ -226,4 +246,40 @@ with col2:
|
|
| 226 |
emotion_placeholder = st.session_state.emotion_placeholder
|
| 227 |
message_placeholder = st.session_state.message_placeholder
|
| 228 |
|
| 229 |
-
st.info("👆 Autorisez l'accès à la webcam dans votre navigateur pour démarrer la détection d'émotions.")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 15 |
def __init__(self):
|
| 16 |
super(EmotionCNN, self).__init__()
|
| 17 |
self.conv_layers = nn.Sequential(
|
|
|
|
| 18 |
nn.Conv2d(1, 32, 3, padding=1),
|
| 19 |
nn.BatchNorm2d(32),
|
| 20 |
nn.ReLU(),
|
|
|
|
| 23 |
nn.ReLU(),
|
| 24 |
nn.MaxPool2d(2),
|
| 25 |
nn.Dropout2d(0.25),
|
|
|
|
|
|
|
| 26 |
nn.Conv2d(32, 64, 3, padding=1),
|
| 27 |
nn.BatchNorm2d(64),
|
| 28 |
nn.ReLU(),
|
|
|
|
| 31 |
nn.ReLU(),
|
| 32 |
nn.MaxPool2d(2),
|
| 33 |
nn.Dropout2d(0.25),
|
|
|
|
|
|
|
| 34 |
nn.Conv2d(64, 128, 3, padding=1),
|
| 35 |
nn.BatchNorm2d(128),
|
| 36 |
nn.ReLU(),
|
|
|
|
| 40 |
nn.MaxPool2d(2),
|
| 41 |
nn.Dropout2d(0.25)
|
| 42 |
)
|
|
|
|
| 43 |
self.fc_layers = nn.Sequential(
|
| 44 |
nn.Linear(128 * 6 * 6, 512),
|
| 45 |
nn.ReLU(),
|
|
|
|
| 56 |
x = self.fc_layers(x)
|
| 57 |
return x
|
| 58 |
|
| 59 |
+
# Dictionnaire des émotions
|
| 60 |
emotion_dict = {
|
| 61 |
0: {"name": "Colère", "message": "Respirez profondément et prenez un moment pour vous calmer."},
|
| 62 |
1: {"name": "Mépris", "message": "Essayez de voir les choses d'un autre point de vue."},
|
|
|
|
| 123 |
def load_model():
|
| 124 |
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
| 125 |
model = EmotionCNN().to(device)
|
| 126 |
+
try:
|
| 127 |
+
model.load_state_dict(torch.load("cnn_emotion_model.pth", map_location=device))
|
| 128 |
+
model.eval()
|
| 129 |
+
except Exception as e:
|
| 130 |
+
st.error(f"Erreur lors du chargement du modèle : {str(e)}")
|
| 131 |
+
st.stop()
|
| 132 |
return model, device
|
| 133 |
|
| 134 |
# Chargement du modèle
|
|
|
|
| 144 |
|
| 145 |
# Chargement du classificateur Haar pour la détection de visage
|
| 146 |
face_cascade = cv2.CascadeClassifier(cv2.data.haarcascades + 'haarcascade_frontalface_default.xml')
|
| 147 |
+
if face_cascade.empty():
|
| 148 |
+
st.error("Erreur : Impossible de charger le classificateur Haar pour la détection de visage.")
|
| 149 |
+
st.stop()
|
| 150 |
|
| 151 |
def detect_faces(frame):
|
| 152 |
gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
|
| 153 |
faces = face_cascade.detectMultiScale(gray, 1.1, 4)
|
| 154 |
return faces
|
| 155 |
|
| 156 |
+
# Configuration RTC avec plusieurs STUN et TURN
|
| 157 |
RTC_CONFIGURATION = RTCConfiguration({
|
| 158 |
+
"iceServers": [
|
| 159 |
+
{"urls": "stun:stun.l.google.com:19302"},
|
| 160 |
+
{"urls": "stun:stun1.l.google.com:19302"},
|
| 161 |
+
{"urls": "stun:stun2.l.google.com:19302"},
|
| 162 |
+
{"urls": "stun:stun3.l.google.com:19302"},
|
| 163 |
+
{"urls": "stun:stun4.l.google.com:19302"},
|
| 164 |
+
{"urls": "stun:stun.stunprotocol.org:3478"},
|
| 165 |
+
# Exemple de configuration TURN (remplacez par vos propres identifiants si disponible)
|
| 166 |
+
{
|
| 167 |
+
"urls": "turn:your-turn-server.example.com:3478",
|
| 168 |
+
"username": "your-username",
|
| 169 |
+
"credential": "your-password"
|
| 170 |
+
}
|
| 171 |
+
]
|
| 172 |
})
|
| 173 |
|
| 174 |
# Classe pour traiter les frames vidéo
|
|
|
|
| 183 |
self.message_placeholder = st.session_state.get('message_placeholder')
|
| 184 |
|
| 185 |
def recv(self, frame):
|
| 186 |
+
try:
|
| 187 |
+
img = frame.to_ndarray(format="bgr24")
|
| 188 |
+
faces = detect_faces(img)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 189 |
|
| 190 |
+
for (x, y, w, h) in faces:
|
| 191 |
+
cv2.rectangle(img, (x, y), (x+w, y+h), (0, 255, 0), 2)
|
| 192 |
+
face_img = img[y:y+h, x:x+w]
|
| 193 |
+
pil_img = Image.fromarray(cv2.cvtColor(face_img, cv2.COLOR_BGR2RGB))
|
| 194 |
+
img_tensor = self.transform(pil_img).unsqueeze(0).to(self.device)
|
| 195 |
+
with torch.no_grad():
|
| 196 |
+
output = self.model(img_tensor)
|
| 197 |
+
_, predicted = torch.max(output, 1)
|
| 198 |
+
emotion_idx = predicted.item()
|
| 199 |
+
emotion_name = self.emotion_dict[emotion_idx]["name"]
|
| 200 |
+
cv2.putText(img, emotion_name, (x, y-10),
|
| 201 |
+
cv2.FONT_HERSHEY_SIMPLEX, 0.9, (0, 255, 0), 2)
|
| 202 |
+
|
| 203 |
+
# Mettre à jour les placeholders
|
| 204 |
+
if self.emotion_placeholder:
|
| 205 |
+
self.emotion_placeholder.markdown(f"""
|
| 206 |
+
<div class="emotion-box">
|
| 207 |
+
<div class="emotion-title">{emotion_name}</div>
|
| 208 |
+
</div>
|
| 209 |
+
""", unsafe_allow_html=True)
|
| 210 |
+
|
| 211 |
+
if self.message_placeholder:
|
| 212 |
+
self.message_placeholder.markdown(f"""
|
| 213 |
+
<div class="emotion-box">
|
| 214 |
+
<div class="emotion-message">{self.emotion_dict[emotion_idx]["message"]}</div>
|
| 215 |
+
</div>
|
| 216 |
+
""", unsafe_allow_html=True)
|
| 217 |
+
return av.VideoFrame.from_ndarray(img, format="bgr24")
|
| 218 |
+
except Exception as e:
|
| 219 |
+
st.error(f"Erreur lors du traitement de la frame : {str(e)}")
|
| 220 |
+
return frame
|
| 221 |
|
| 222 |
+
# Configuration de l'interface
|
| 223 |
col1, col2 = st.columns([2, 1])
|
| 224 |
|
| 225 |
with col1:
|
| 226 |
st.markdown("### 📹 Flux Vidéo")
|
| 227 |
+
try:
|
| 228 |
+
webrtc_ctx = webrtc_streamer(
|
| 229 |
+
key="emotion-detection",
|
| 230 |
+
rtc_configuration=RTC_CONFIGURATION,
|
| 231 |
+
video_processor_factory=VideoProcessor,
|
| 232 |
+
media_stream_constraints={"video": True, "audio": False},
|
| 233 |
+
async_processing=True
|
| 234 |
+
)
|
| 235 |
+
except Exception as e:
|
| 236 |
+
st.error(f"Erreur lors de l'initialisation de WebRTC : {str(e)}")
|
| 237 |
+
st.warning("Vérifiez votre connexion réseau ou les paramètres STUN/TURN.")
|
| 238 |
|
| 239 |
with col2:
|
| 240 |
st.markdown("### 😊 Émotion Détectée")
|
|
|
|
| 241 |
if 'emotion_placeholder' not in st.session_state:
|
| 242 |
st.session_state.emotion_placeholder = st.empty()
|
| 243 |
if 'message_placeholder' not in st.session_state:
|
|
|
|
| 246 |
emotion_placeholder = st.session_state.emotion_placeholder
|
| 247 |
message_placeholder = st.session_state.message_placeholder
|
| 248 |
|
| 249 |
+
st.info("👆 Autorisez l'accès à la webcam dans votre navigateur pour démarrer la détection d'émotions.")
|
| 250 |
+
st.warning("Si la connexion échoue, vérifiez votre réseau ou configurez un serveur TURN pour WebRTC.")
|
| 251 |
+
|
| 252 |
+
# Option de téléchargement d'image comme solution de secours
|
| 253 |
+
st.markdown("### 📷 Ou téléchargez une image")
|
| 254 |
+
uploaded_file = st.file_uploader("Choisissez une image...", type=["jpg", "jpeg", "png"])
|
| 255 |
+
if uploaded_file is not None:
|
| 256 |
+
image = Image.open(uploaded_file)
|
| 257 |
+
frame = np.array(image)
|
| 258 |
+
faces = detect_faces(frame)
|
| 259 |
+
|
| 260 |
+
for (x, y, w, h) in faces:
|
| 261 |
+
cv2.rectangle(frame, (x, y), (x+w, y+h), (0, 255, 0), 2)
|
| 262 |
+
face_img = frame[y:y+h, x:x+w]
|
| 263 |
+
pil_img = Image.fromarray(cv2.cvtColor(face_img, cv2.COLOR_BGR2RGB))
|
| 264 |
+
img_tensor = transform(pil_img).unsqueeze(0).to(device)
|
| 265 |
+
with torch.no_grad():
|
| 266 |
+
output = model(img_tensor)
|
| 267 |
+
_, predicted = torch.max(output, 1)
|
| 268 |
+
emotion_idx = predicted.item()
|
| 269 |
+
emotion_name = emotion_dict[emotion_idx]["name"]
|
| 270 |
+
cv2.putText(frame, emotion_name, (x, y-10),
|
| 271 |
+
cv2.FONT_HERSHEY_SIMPLEX, 0.9, (0, 255, 0), 2)
|
| 272 |
+
|
| 273 |
+
emotion_placeholder.markdown(f"""
|
| 274 |
+
<div class="emotion-box">
|
| 275 |
+
<div class="emotion-title">{emotion_name}</div>
|
| 276 |
+
</div>
|
| 277 |
+
""", unsafe_allow_html=True)
|
| 278 |
+
|
| 279 |
+
message_placeholder.markdown(f"""
|
| 280 |
+
<div class="emotion-box">
|
| 281 |
+
<div class="emotion-message">{emotion_dict[emotion_idx]["message"]}</div>
|
| 282 |
+
</div>
|
| 283 |
+
""", unsafe_allow_html=True)
|
| 284 |
+
|
| 285 |
+
st.image(cv2.cvtColor(frame, cv2.COLOR_BGR2RGB))
|