model fuse

detect.py

@@ -21,6 +21,8 @@ def detect(save_img=False):
     google_utils.attempt_download(weights)
     model = torch.load(weights, map_location=device)['model']
     # torch.save(torch.load(weights, map_location=device), weights)  # update model if SourceChangeWarning
+    # model.fuse()
+    model.to(device).eval()
 
     # Second-stage classifier
     classify = False
@@ -29,12 +31,6 @@ def detect(save_img=False):
         modelc.load_state_dict(torch.load('weights/resnet101.pt', map_location=device)['model'])  # load weights
         modelc.to(device).eval()
 
-    # Eval mode
-    model.to(device).eval()
-
-    # Fuse Conv2d + BatchNorm2d layers
-    # model.fuse()
-
     # Half precision
     half = half and device.type != 'cpu'  # half precision only supported on CUDA
     if half: