Update

app.css

@@ -36,6 +36,10 @@ div.stat {
     height: 286px;
 }
 
+div.settings-wrapper {
+    display: none;
+}
+
 .submit {
     display: inline-block;
     padding: 10px 20px;

app/app_utils.py

@@ -14,7 +14,7 @@ import cv2
 # Importing necessary components for the Gradio app
 from app.model import pth_model_static, pth_model_dynamic, pth_processing
 from app.face_utils import get_box, display_info
-from app.config import DICT_EMO
+from app.config import DICT_EMO, config_data
 from app.plot import statistics_plot
 
 
@@ -43,9 +43,9 @@ def preprocess_image_and_predict(inp):
             for fl in results.multi_face_landmarks:
                 startX, startY, endX, endY = get_box(fl, w, h)
                 cur_face = inp[startY:endY, startX:endX]
-                cur_face_n = pth_processing(Image.fromarray(cur_face))
+                cur_face_n = pth_processing(Image.fromarray(cur_face)).to(config_data.DEVICE)
                 prediction = (
-                    torch.nn.functional.softmax(pth_model_static(cur_face_n), dim=1)
+                    torch.nn.functional.softmax(pth_model_static(cur_face_n), dim=1).cpu()
                     .detach()
                     .numpy()[0]
                 )
@@ -92,8 +92,8 @@ def preprocess_video_and_predict(video):
                     cur_face = frame_copy[startY:endY, startX: endX]
 
                     if (count_frame-1)%5 == 0:
-                        cur_face_copy = pth_processing(Image.fromarray(cur_face))
-                        features = torch.nn.functional.relu(pth_model_static.extract_features(cur_face_copy)).detach().numpy()
+                        cur_face_copy = pth_processing(Image.fromarray(cur_face)).to(config_data.DEVICE)
+                        features = torch.nn.functional.relu(pth_model_static.extract_features(cur_face_copy)).cpu().detach().numpy()
         
                         if len(lstm_features) == 0:
                             lstm_features = [features]*10
@@ -101,8 +101,8 @@ def preprocess_video_and_predict(video):
                             lstm_features = lstm_features[1:] + [features]
 
                         lstm_f = torch.from_numpy(np.vstack(lstm_features))
-                        lstm_f = torch.unsqueeze(lstm_f, 0)
-                        output = pth_model_dynamic(lstm_f).detach().numpy()
+                        lstm_f = torch.unsqueeze(lstm_f, 0).to(config_data.DEVICE)
+                        output = pth_model_dynamic(lstm_f).cpu().detach().numpy()
                         last_output = output
                     else:
                         if last_output is not None:

app/model.py

@@ -27,9 +27,9 @@ def load_model(model_url, model_path):
         return None
 
 
-pth_model_static = load_model(config_data.model_static_url, config_data.model_static_path)
+pth_model_static = load_model(config_data.model_static_url, config_data.model_static_path).to(config_data.DEVICE)
 
-pth_model_dynamic = load_model(config_data.model_dynamic_url, config_data.model_dynamic_path)
+pth_model_dynamic = load_model(config_data.model_dynamic_url, config_data.model_dynamic_path).to(config_data.DEVICE)
 
 
 

config.toml

@@ -1,4 +1,5 @@
 APP_VERSION = "0.2.0"
+DEVICE = "cpu"
 
 [model_static]
 url = "https://huggingface.co/ElenaRyumina/face_emotion_recognition/resolve/main/FER_static_ResNet50_AffectNet.pth"