diff --git a/src/membrain_seg/annotations/extract_patch_cli.py b/src/membrain_seg/annotations/extract_patch_cli.py
index c984642..339e476 100644
--- a/src/membrain_seg/annotations/extract_patch_cli.py
+++ b/src/membrain_seg/annotations/extract_patch_cli.py
@@ -84,7 +84,7 @@ def extract_patches(
     pad_value = 2.0  # Currently still hard-coded because other values not
     # compatible with training routine yet.
     if coords_file is not None:
-        coords = get_csv_data(csv_path=coords_file)
+        coords = np.array(get_csv_data(csv_path=coords_file), dtype=int)
     else:
         coords = [np.array((x, y, z))]
     _extract_patches(
diff --git a/src/membrain_seg/annotations/merge_corrections.py b/src/membrain_seg/annotations/merge_corrections.py
index 951083e..8286633 100644
--- a/src/membrain_seg/annotations/merge_corrections.py
+++ b/src/membrain_seg/annotations/merge_corrections.py
@@ -105,8 +105,13 @@ def convert_single_nrrd_files(labels_dir, corrections_dir, out_dir):
     for label_file in os.listdir(labels_dir):
         if not os.path.isfile(os.path.join(labels_dir, label_file)):
             continue
+        print("")
         print("Finding correction files for", label_file)
-        token = os.path.splitext(label_file)[0]
+        # token = os.path.splitext(label_file)[0]
+        if label_file.endswith(".nii.gz"):
+            token = label_file[:-7]
+        elif label_file.endswith(".nrrd"):
+            token = label_file[:-5]
         found_flag = 0
         for filename in os.listdir(corrections_dir):
             if not os.path.isdir(os.path.join(corrections_dir, filename)):
@@ -120,5 +125,5 @@ def convert_single_nrrd_files(labels_dir, corrections_dir, out_dir):
                 print("Storing corrections in", out_file)
                 write_nifti(out_file, merged_corrections)
                 found_flag = 1
-            if found_flag == 0:
-                print("No corrections folder found for patch", token)
+        if found_flag == 0:
+            print("No corrections folder found for patch", token)
diff --git a/src/membrain_seg/segmentation/dataloading/data_utils.py b/src/membrain_seg/segmentation/dataloading/data_utils.py
index e307784..28109d9 100644
--- a/src/membrain_seg/segmentation/dataloading/data_utils.py
+++ b/src/membrain_seg/segmentation/dataloading/data_utils.py
@@ -118,7 +118,7 @@ def load_data_for_inference(data_path: str, transforms: Callable, device: device
     new_data = transforms(new_data)
     new_data = new_data.unsqueeze(0)  # Add batch dimension
     new_data = new_data.to(device)
-    return new_data, tomogram.header
+    return new_data, tomogram.header, tomogram.voxel_size
 
 
 def store_segmented_tomograms(
@@ -130,6 +130,7 @@ def store_segmented_tomograms(
     store_connected_components: bool = False,
     connected_component_thres: int = None,
     mrc_header: np.recarray = None,
+    voxel_size: float = None,
 ) -> None:
     """
     Helper function for storing output segmentations.
@@ -159,6 +160,9 @@ def store_segmented_tomograms(
         If given, the mrc header will be used to retain header information
         from another tomogram. This way, pixel sizes and other header
         information is not lost.
+    voxel_size: float, optional
+        If given, this will be the voxel size stored in the header of the
+        output segmentation.
     """
     # Create out directory if it doesn't exist yet
     make_directory_if_not_exists(out_folder)
@@ -170,7 +174,9 @@ def store_segmented_tomograms(
         out_file = os.path.join(
             out_folder, os.path.basename(orig_data_path)[:-4] + "_scores.mrc"
         )
-        out_tomo = Tomogram(data=predictions_np, header=mrc_header)
+        out_tomo = Tomogram(
+            data=predictions_np, header=mrc_header, voxel_size=voxel_size
+        )
         store_tomogram(out_file, out_tomo)
     predictions_np_thres = predictions.squeeze(0).squeeze(0).cpu().numpy() > 0.0
     out_file_thres = os.path.join(
@@ -181,7 +187,9 @@ def store_segmented_tomograms(
         predictions_np_thres = connected_components(
             predictions_np_thres, size_thres=connected_component_thres
         )
-    out_tomo = Tomogram(data=predictions_np_thres, header=mrc_header)
+    out_tomo = Tomogram(
+        data=predictions_np_thres, header=mrc_header, voxel_size=voxel_size
+    )
     store_tomogram(out_file_thres, out_tomo)
     print("MemBrain has finished segmenting your tomogram.")
     return out_file_thres
@@ -334,7 +342,7 @@ def convert_dtype(tomogram: np.ndarray) -> np.ndarray:
     if (
         tomogram.min() >= np.finfo("float16").min
         and tomogram.max() <= np.finfo("float16").max
-    ):
+    ) and np.allclose(tomogram, tomogram.astype("float16")):
         return tomogram.astype("float16")
     elif (
         tomogram.min() >= np.finfo("float32").min
@@ -368,33 +376,22 @@ def store_tomogram(
         if isinstance(tomogram, Tomogram):
             data = tomogram.data
             header = tomogram.header
+            if voxel_size is None:
+                voxel_size = tomogram.voxel_size
         else:
             data = tomogram
             header = None
-        data = convert_dtype(data)
-        data = np.transpose(data, (2, 1, 0))
-        dtype_mode = _dtype_to_mode[data.dtype]
-        out_mrc.set_data(data)
+
         if header is not None:
             attributes = header.dtype.names
             for attr in attributes:
-                # skip density and shape attribues
-                if attr in [
-                    "mode",
-                    "dmean",
-                    "dmin",
-                    "dmax",
-                    "rms",
-                    "nx",
-                    "ny",
-                    "nz",
-                    "mx",
-                    "my",
-                    "mz",
-                ]:
+                if attr not in ["nlabl", "label"]:
                     continue
                 setattr(out_mrc.header, attr, getattr(header, attr))
-            out_mrc.header.mode = dtype_mode
+
+        data = convert_dtype(data)
+        data = np.transpose(data, (2, 1, 0))
+        out_mrc.set_data(data)
         if voxel_size is not None:
             out_mrc.voxel_size = voxel_size
 
diff --git a/src/membrain_seg/segmentation/segment.py b/src/membrain_seg/segmentation/segment.py
index 513a806..5d34b88 100644
--- a/src/membrain_seg/segmentation/segment.py
+++ b/src/membrain_seg/segmentation/segment.py
@@ -20,7 +20,7 @@ def segment(
     sw_roi_size=160,
     store_connected_components=False,
     connected_component_thres=None,
-    test_time_augmentation=True
+    test_time_augmentation=True,
 ):
     """
     Segment tomograms using a trained model.
@@ -81,7 +81,7 @@ def segment(
     # Preprocess the new data
     new_data_path = tomogram_path
     transforms = get_prediction_transforms()
-    new_data, mrc_header = load_data_for_inference(
+    new_data, mrc_header, voxel_size = load_data_for_inference(
         new_data_path, transforms, device=torch.device("cpu")
     )
     new_data = new_data.to(torch.float32)
@@ -106,7 +106,7 @@ def segment(
     # Perform test time augmentation (8-fold mirroring)
     predictions = torch.zeros_like(new_data)
     print("Performing 8-fold test-time augmentation.")
-    for m in range((8 if test_time_augmentation else 1)):
+    for m in range(8 if test_time_augmentation else 1):
         with torch.no_grad():
             with torch.cuda.amp.autocast():
                 predictions += (
@@ -132,5 +132,6 @@ def segment(
         store_connected_components=store_connected_components,
         connected_component_thres=connected_component_thres,
         mrc_header=mrc_header,
+        voxel_size=voxel_size,
     )
     return segmentation_file