change cfg loading to fix cpu mode

cutie/config/config.py

@@ -0,0 +1,4 @@
+from hydra import compose, initialize
+
+initialize(version_base='1.3.2', config_path="", job_name="gui")
+global_config = compose(config_name="gui_config")

cutie_roto.py

@@ -13,11 +13,11 @@
     sys.exit("Please execute \"install_pytorch.bat\" to install Pytorch, then try again.")
 
 from omegaconf import open_dict
-from hydra import compose, initialize
 import logging
 from PySide6.QtWidgets import QApplication, QDialog
 import qdarktheme
 
+from cutie.config.config import global_config
 from gui.main_controller import MainController
 from gui.launcher_gui import Launcher_Dialog
 
@@ -54,9 +54,8 @@ def get_arguments():
 if __name__ in "__main__":
     log = logging.getLogger()
 
-    # getting hydra's config without using its decorator
-    initialize(version_base='1.3.2', config_path="cutie/config", job_name="gui")
-    cfg = compose(config_name="gui_config")
+    # get the config
+    cfg = global_config
 
     # input arguments
     args = get_arguments()

gui/interactive_utils.py

@@ -6,6 +6,7 @@
 import torch
 import torch.nn.functional as F
 from cutie.utils.palette import davis_palette
+from cutie.config.config import global_config
 
 
 def image_to_torch(frame: np.ndarray, device: str = 'cuda'):
@@ -26,22 +27,19 @@ def index_numpy_to_one_hot_torch(mask: np.ndarray, num_classes: int):
     return F.one_hot(mask, num_classes=num_classes).permute(2, 0, 1).float()
 
 
-
-# Some constants for visualization
-"""
-try:
-    if torch.cuda.is_available():
-        device = torch.device("cuda")
-    elif torch.backends.mps.is_available():
-        device = torch.device("mps")
-    else:
-        device = torch.device("cpu")
-except:
+# set torch device for interactice segmentation
+cfg = global_config
+if cfg.force_cpu:
     device = torch.device("cpu")
-"""
-# get existing device instead of detecting again
-device = torch.cuda.current_device()
+elif torch.cuda.is_available():
+    device = torch.device("cuda")
+elif torch.backends.mps.is_available():
+    device = torch.device("mps")
+else:
+    device = torch.device("cpu")
+#print(f'Using click device: {device}')
 
+# Some constants for visualization
 color_map_np = np.frombuffer(davis_palette, dtype=np.uint8).reshape(-1, 3).copy()
 # scales for better visualization
 color_map_np = (color_map_np.astype(np.float32) * 1.5).clip(0, 255).astype(np.uint8)