Semi supervised learning pipeline notebook

models/model_32offjns/requirements.txt

@@ -0,0 +1,3 @@
+ultralytics
+opencv-python
+ipykernel

models/model_32offjns/src/architecture.py

@@ -41,13 +41,13 @@ def __init__(self, lookback):
     self.bottleneck = self.conv_block(512, 1024)
 
     # Expansive path
-    self.upconv4 = nn.ConvTranspose2d(1024, 512, kernel_size=2, stride=1)
+    self.upconv4 = nn.ConvTranspose2d(1024, 512, kernel_size=2, stride=2)
     self.dec4 = self.conv_block(1024, 512)
-    self.upconv3 = nn.ConvTranspose2d(512, 256, kernel_size=2, stride=1)
+    self.upconv3 = nn.ConvTranspose2d(512, 256, kernel_size=2, stride=2)
     self.dec3 = self.conv_block(512, 256)
-    self.upconv2 = nn.ConvTranspose2d(256, 128, kernel_size=2, stride=1)
+    self.upconv2 = nn.ConvTranspose2d(256, 128, kernel_size=2, stride=2)
     self.dec2 = self.conv_block(256, 128)
-    self.upconv1 = nn.ConvTranspose2d(128, 64, kernel_size=2, stride=1)
+    self.upconv1 = nn.ConvTranspose2d(128, 64, kernel_size=2, stride=2)
     self.dec1 = self.conv_block(128, 64)
 
     # Final layer

models/model_32offjns/src/dataset.py

@@ -114,10 +114,30 @@ def __getitem__(self, idx):
     label_drivable_area_dir = f"{self.dataset_dir}/{self.data[idx]['dataset']}/label/drivable_area/{data_idx_str}.jpg"
     label_cones_dir = f"{self.dataset_dir}/{self.data[idx]['dataset']}/label/cones/{data_idx_str}.jpg"
 
-    label_background = torch.tensor(cv2.imread(label_background_dir, cv2.IMREAD_GRAYSCALE), device=self.device, dtype=torch.float32)
-    label_lane_lines = torch.tensor(cv2.imread(label_lane_lines_dir, cv2.IMREAD_GRAYSCALE), device=self.device, dtype=torch.float32)
-    label_drivable_area = torch.tensor(cv2.imread(label_drivable_area_dir, cv2.IMREAD_GRAYSCALE), device=self.device, dtype=torch.float32)
-    label_cones = torch.tensor(cv2.imread(label_cones_dir, cv2.IMREAD_GRAYSCALE), device=self.device, dtype=torch.float32)
+    # label_background = torch.tensor(cv2.imread(label_background_dir, cv2.IMREAD_GRAYSCALE), device=self.device, dtype=torch.float32)
+    # label_lane_lines = torch.tensor(cv2.imread(label_lane_lines_dir, cv2.IMREAD_GRAYSCALE), device=self.device, dtype=torch.float32)
+    # label_drivable_area = torch.tensor(cv2.imread(label_drivable_area_dir, cv2.IMREAD_GRAYSCALE), device=self.device, dtype=torch.float32)
+    # label_cones = torch.tensor(cv2.imread(label_cones_dir, cv2.IMREAD_GRAYSCALE), device=self.device, dtype=torch.float32)
+    label_background = torch.tensor(
+        cv2.imread(label_background_dir, cv2.IMREAD_GRAYSCALE) if cv2.imread(label_background_dir, cv2.IMREAD_GRAYSCALE) is not None else np.zeros((384, 640), dtype=np.uint8),
+        device=self.device,
+        dtype=torch.float32
+    )
+    label_lane_lines = torch.tensor(
+        cv2.imread(label_lane_lines_dir, cv2.IMREAD_GRAYSCALE) if cv2.imread(label_lane_lines_dir, cv2.IMREAD_GRAYSCALE) is not None else np.zeros((384, 640), dtype=np.uint8),
+        device=self.device,
+        dtype=torch.float32
+    )
+    label_drivable_area = torch.tensor(
+        cv2.imread(label_drivable_area_dir, cv2.IMREAD_GRAYSCALE) if cv2.imread(label_drivable_area_dir, cv2.IMREAD_GRAYSCALE) is not None else np.zeros((384, 640), dtype=np.uint8),
+        device=self.device,
+        dtype=torch.float32
+    )
+    label_cones = torch.tensor(
+        cv2.imread(label_cones_dir, cv2.IMREAD_GRAYSCALE) if cv2.imread(label_cones_dir, cv2.IMREAD_GRAYSCALE) is not None else np.zeros((384, 640), dtype=np.uint8),
+        device=self.device,
+        dtype=torch.float32
+    )
 
     label_shape = label_background.shape
     label = torch.zeros((label_shape[0], label_shape[1], 4), device=self.device, dtype=torch.long)

models/model_32offjns/src/methods.py

@@ -49,6 +49,22 @@ def upload_model_weights(model, dbx_access_token, delete=True):
     if delete:
         os.remove(local_model_weights_dir)
 
+def upload_dataset_to_dropbox(dataset_name, dbx_access_token):
+    if dbx_access_token == "":
+        print("Dropbox access token uninitialized. Unable to upload dataset.")
+        return
+    try:
+        dbx = dropbox.Dropbox(dbx_access_token)
+    except:
+        print("Could not connect to Dropbox when attempting to upload dataset.")
+        return
+    # TODO: modify slightly to accomodate the upload of data and labels properly
+    dbx_dataset_dir = f'/UMARV/ComputerVision/ScenePerception/datasets/{dataset_name}'
+    local_dataset_dir = f'{os.getenv("REPO_DIR")}/datasets/{dataset_name}'
+    with open(local_dataset_dir, 'rb') as file:
+        dbx.files_upload(file.read(), dbx_dataset_dir, mode=dropbox.files.WriteMode("overwrite"))
+    print(f"Uploaded dataset \"{dataset_name}\" to Dropbox.")
+
 def download_model_weights(model, dbx_access_token, delete=True):
     if dbx_access_token == "":
         print("Dropbox access token uninitialized. Unable to download model weights.")