Update app.py

app.py

@@ -6,6 +6,41 @@ import torch
 from PIL import Image
 import torchvision.transforms as transforms
 
+
+def load_model():
+    # CNN3D Layer's architecture
+    cnndata = CNNData(in_dim = 1,
+                      n_f =[32,48],
+                      kernel_size=[(5,5,5), (3,3,3)],
+                      activations=[nn.ReLU(),nn.ReLU()],
+                      bns = [True, True],
+                      dropouts = [0, 0],
+                      paddings = [(0,0,0),(0,0,0)],
+                      strides = [(2,2,2),(2,2,2)])
+
+    # Feedforward layer's architecture
+    lindata = LinData(in_dim = conv3D_output_size(cnndata, [30, 256, 342]),
+                      hidden_layers= [256,256,1],
+                      activations=[nn.ReLU(),nn.ReLU(),None],
+                      bns=[False,False,False],
+                      dropouts =[0.2, 0, 0])
+
+    # combined architecture
+    args = NetData(cnndata, lindata)
+
+    # weight file
+    #weight_file = 'cnn3d_epoch_300.pt'
+       
+    # CNN3D model
+    # device = 'cuda' if torch.cuda.is_available() else 'cpu'
+    device = torch.device('cpu')
+    cnn3d = CNN3D(args).to(device)
+    #cnn3d.load_state_dict(torch.load(os.path.join(base_path,'weights',weight_file), map_location=device))
+    cnn3d.eval()
+    #print(cnn3d)
+
+    return cnn3d
+
 def parse_video(video_file):
     """A utility to parse the input videos.
     Reference: https://pyimagesearch.com/2018/11/12/yolo-object-detection-with-opencv/
@@ -46,6 +81,8 @@ def parse_video(video_file):
     return frames
 
 def pil_parser(video_file):
+
+    model = load_model()
     # cv2 parsing
 
     dummy_frames = parse_video(video_file)