Allow user to specify additional features to be returned by Image Dataset when using ReaderHfds

Author: grodino

timm/data/dataset.py

@@ -30,6 +30,7 @@ def __init__(
             input_img_mode='RGB',
             transform=None,
             target_transform=None,
+            additional_features=None,
             **kwargs,
     ):
         if reader is None or isinstance(reader, str):
@@ -38,17 +39,19 @@ def __init__(
                 root=root,
                 split=split,
                 class_map=class_map,
+                additional_features=additional_features,
                 **kwargs,
             )
         self.reader = reader
         self.load_bytes = load_bytes
         self.input_img_mode = input_img_mode
         self.transform = transform
         self.target_transform = target_transform
+        self.additional_features = additional_features
         self._consecutive_errors = 0
 
     def __getitem__(self, index):
-        img, target = self.reader[index]
+        img, target, *features = self.reader[index]
 
         try:
             img = img.read() if self.load_bytes else Image.open(img)
@@ -71,7 +74,10 @@ def __getitem__(self, index):
         elif self.target_transform is not None:
             target = self.target_transform(target)
 
-        return img, target
+        if self.additional_features is None:
+            return img, target
+        else:
+            return img, target, *features
 
     def __len__(self):
         return len(self.reader)

timm/data/readers/reader_factory.py

@@ -19,11 +19,14 @@ def create_reader(
         prefix = name[0]
     name = name[-1]
 
+    # FIXME the additional features are only supported by ReaderHfds for now.
+    additional_features = kwargs.pop("additional_features", None)
+
     # FIXME improve the selection right now just tfds prefix or fallback path, will need options to
     # explicitly select other options shortly
     if prefix == 'hfds':
         from .reader_hfds import ReaderHfds  # defer Hf datasets import
-        reader = ReaderHfds(name=name, root=root, split=split, **kwargs)
+        reader = ReaderHfds(name=name, root=root, split=split, additional_features=additional_features, **kwargs)
     elif prefix == 'hfids':
         from .reader_hfids import ReaderHfids  # defer HF datasets import
         reader = ReaderHfids(name=name, root=root, split=split, **kwargs)

timm/data/readers/reader_hfds.py

@@ -37,6 +37,7 @@ def __init__(
             class_map: dict = None,
             input_key: str = 'image',
             target_key: str = 'label',
+            additional_features: Optional[list[str]] = None,
             download: bool = False,
             trust_remote_code: bool = False
     ):
@@ -65,9 +66,18 @@ def __init__(
         self.split_info = self.dataset.info.splits[split]
         self.num_samples = self.split_info.num_examples
 
+        if isinstance(additional_features, str):
+            self.additional_features = [additional_features]
+        elif isinstance(additional_features, list):
+            self.additional_features = additional_features
+        else:
+            self.additional_features = []
+
     def __getitem__(self, index):
         item = self.dataset[index]
         image = item[self.image_key]
+        features = [item[feat] for feat in self.additional_features]
+
         if 'bytes' in image and image['bytes']:
             image = io.BytesIO(image['bytes'])
         else:
@@ -76,7 +86,8 @@ def __getitem__(self, index):
         label = item[self.label_key]
         if self.remap_class:
             label = self.class_to_idx[label]
-        return image, label
+
+        return image, label, *features
 
     def __len__(self):
         return len(self.dataset)