DeepLabCut · kimmyjuarez · Oct 19, 2025 · Nov 6, 2025 · Nov 8, 2025 · Nov 13, 2025
diff --git a/deeplabcut/pose_estimation_pytorch/data/transforms.py b/deeplabcut/pose_estimation_pytorch/data/transforms.py
@@ -22,9 +22,25 @@
 from scipy.stats import truncnorm
 
 
-def build_transforms(augmentations: dict) -> A.BaseCompose:
-    transforms = []
 
+def transforms_legacy(augmentations):
+
+    """
+    Apply DeepLabCut original augmentation methods.
+
+    This function preserves DLC's default augmentation implementations
+    for backward compatibility.
+
+    Args:
+        augmentations: dictionary of augmentations from config file
+
+    Returns:
+
+        The list of augmentation transforms
+
+    """
+
+    transforms = []
     if resize_aug := augmentations.get("resize", False):
         transforms += build_resize_transforms(resize_aug)
 
@@ -121,6 +137,7 @@ def build_transforms(augmentations: dict) -> A.BaseCompose:
         transforms.append(ElasticTransform(sigma=5, p=0.5))
     if augmentations.get("grayscale", False):
         transforms.append(Grayscale(alpha=(0.5, 1.0)))
+
     if noise := augmentations.get("gaussian_noise", False):
         # TODO inherit custom gaussian transform to support per_channel = 0.5
         if not isinstance(noise, (int, float)):
@@ -146,6 +163,54 @@ def build_transforms(augmentations: dict) -> A.BaseCompose:
     if augmentations.get("scale_to_unit_range"):
         transforms.append(ScaleToUnitRange())
 
+    return transforms
+
+
+
+def transform_new(augmentations: list[dict[str, Any]]) -> list[A.BasicTransform]:
+
+    """
+    Creates augmentation using Albumentation library with hybrid support.
+
+    Args:
+        augmentations: the augmentations from "transform" in list format
+
+    Raises:
+        Warnings:
+            Augmentation method must exist in augmentations.
+            Augmentation must exist in the Albumentation library.
+
+    Returns:
+        The list of augmentation transforms
+
+    """
+
+    custom_augmentations =['auto_padding', 'gaussian_noise', 'motion_blur', 'normalize_images']
+
+    transforms =[]
+    for aug in augmentations:
+        method = aug.get('augmentation', "")
+        if method == "":
+            warnings.warn("No augmentation method specified. Skipping this transform.")
+            continue
+
+        args = {key:value for key, value in aug.items() if key != 'augmentation'}
+        if method.lower() in custom_augmentations:
+                add_aug = transforms_legacy({method.lower(): args})
+                transforms.extend(add_aug)
+        elif hasattr(A, method):
+            func = getattr(A,method)
+            transforms.append(func(**args))
+        else:
+            warnings.warn(f"Albumentations has no method named {method}. Skipping this transform.")
+
+    return transforms
+
+
+def build_transforms(augmentations: dict) -> A.BaseCompose:
+    transforms = transforms_legacy(augmentations)
+    transforms = transforms + transform_new(augmentations.get("transform", []))
+
     return A.Compose(
         transforms,
         keypoint_params=A.KeypointParams(
@@ -155,6 +220,7 @@ def build_transforms(augmentations: dict) -> A.BaseCompose:
     )
 
 
+
 def build_auto_padding(
     min_height: int | None = None,
     min_width: int | None = None,