Ucf101 dataset: add S3 version (non default yet).

PiperOrigin-RevId: 282941574

Author: pierrot0

tensorflow_datasets/video/ucf101.py

@@ -79,6 +79,14 @@ def __init__(self, split_number, width=None, height=None, **kwargs):
     self.split_number = split_number
 
 
+_VERSION = tfds.core.Version(
+    '1.0.0', experiments={tfds.core.Experiment.S3: False})
+_SUPPORTED_VERSIONS = [
+    tfds.core.Version(
+        '2.0.0', 'New split API (https://tensorflow.org/datasets/splits)'),
+]
+
+
 class Ucf101(tfds.core.GeneratorBasedBuilder):
   """Ucf101 action recognition dataset.
 
@@ -93,35 +101,35 @@ class Ucf101(tfds.core.GeneratorBasedBuilder):
           width=256,
           height=256,
           split_number=1,
-          version=tfds.core.Version(
-              '1.0.0', experiments={tfds.core.Experiment.S3: False}),
+          version=_VERSION,
+          supported_versions=_SUPPORTED_VERSIONS,
       ),
       Ucf101Config(
           name='ucf101_1',
           description='UCF with the action recognition split #1.',
           width=None,
           height=None,
           split_number=1,
-          version=tfds.core.Version(
-              '1.0.0', experiments={tfds.core.Experiment.S3: False}),
+          version=_VERSION,
+          supported_versions=_SUPPORTED_VERSIONS,
       ),
       Ucf101Config(
           name='ucf101_2',
           description='UCF with the action recognition split #2.',
           width=None,
           height=None,
           split_number=2,
-          version=tfds.core.Version(
-              '1.0.0', experiments={tfds.core.Experiment.S3: False}),
+          version=_VERSION,
+          supported_versions=_SUPPORTED_VERSIONS,
       ),
       Ucf101Config(
           name='ucf101_3',
           description='UCF with the action recognition split #3.',
           width=None,
           height=None,
           split_number=3,
-          version=tfds.core.Version(
-              '1.0.0', experiments={tfds.core.Experiment.S3: False}),
+          version=_VERSION,
+          supported_versions=_SUPPORTED_VERSIONS,
       ),
   ]
 
@@ -186,7 +194,7 @@ def _generate_examples(self, videos_dir, splits_dir, data_list):
     data_list_path_path = os.path.join(splits_dir, data_list)
     with tf.io.gfile.GFile(data_list_path_path, 'r') as data_list_file:
       labels_and_paths = data_list_file.readlines()
-    for label_and_path in labels_and_paths:
+    for label_and_path in sorted(labels_and_paths):
       # The train splits contain not only the filename, but also a digit
       # encoding the label separated by a space, which we ignore.
       label_and_path = label_and_path.strip().split(' ')[0]
@@ -198,4 +206,4 @@ def _generate_examples(self, videos_dir, splits_dir, data_list):
         logging.error('Example %s not found', video_path)
         continue
       # We extract the label from the filename.
-      yield {'video': video_path, 'label': label}
+      yield path, {'video': video_path, 'label': label}

tensorflow_datasets/video/ucf101_test.py

@@ -54,5 +54,10 @@ def _assertAsDataset(self, builder):
                      {'test': {'Archery': 1, 'Nunchucks': 1},
                       'train': {'Archery': 1, 'Nunchucks': 2}})
 
+
+class Ucf101S3Test(Ucf101Test):
+  VERSION = 'experimental_latest'
+
+
 if __name__ == '__main__':
   testing.test_main()