Add v2.0.0 Cars Dataset to TFDS, now with test labels.

PiperOrigin-RevId: 282351301

Author: jpuigcerver

tensorflow_datasets/image/cars196.py

@@ -133,7 +133,7 @@
     'Volkswagen Golf Hatchback 2012', 'Volkswagen Golf Hatchback 1991',
     'Volkswagen Beetle Hatchback 2012', 'Volvo C30 Hatchback 2012',
     'Volvo 240 Sedan 1993', 'Volvo XC90 SUV 2007',
-    'smart fortwo Convertible 2012'
+    'smart fortwo Convertible 2012',
 ]
 
 _CITATION = """\
@@ -152,17 +152,21 @@
 class Cars196(tfds.core.GeneratorBasedBuilder):
   """Car Images dataset."""
 
-  VERSION = tfds.core.Version('1.1.0')
+  VERSION = tfds.core.Version('2.0.0')
+
+  SUPPORTED_VERSIONS = [
+      tfds.core.Version('1.1.0'),  # Version with no test labels.
+  ]
 
   def _info(self):
     """Define the dataset info."""
-
+    names = _NAMES + ['test'] if self.version < '2.0.0' else _NAMES
     return tfds.core.DatasetInfo(
         builder=self,
         description=(_DESCRIPTION),
         features=tfds.features.FeaturesDict({
             'image': tfds.features.Image(),
-            'label': tfds.features.ClassLabel(names=_NAMES),
+            'label': tfds.features.ClassLabel(names=names),
             'bbox': tfds.features.BBoxFeature(),
         }),
         supervised_keys=('image', 'label'),
@@ -175,6 +179,8 @@ def _split_generators(self, dl_manager):
         'train': urllib.parse.urljoin(_URL, 'cars_train.tgz'),
         'test': urllib.parse.urljoin(_URL, 'cars_test.tgz'),
         'extra': _EXTRA_URL,
+        'test_annos':
+            urllib.parse.urljoin(_URL, 'cars_test_annos_withlabels.mat'),
     })
 
     return [
@@ -198,9 +204,7 @@ def _split_generators(self, dl_manager):
                     'test',
                 'data_dir_path':
                     os.path.join(output_files['test'], 'cars_test'),
-                'data_annotations_path':
-                    os.path.join(output_files['extra'],
-                                 os.path.join('devkit', 'cars_test_annos.mat')),
+                'data_annotations_path': output_files['test_annos'],
             },
         ),
     ]
@@ -215,9 +219,10 @@ def _generate_examples(self, split_name, data_dir_path,
       mat = tfds.core.lazy_imports.scipy.io.loadmat(f)
     for example in mat['annotations'][0]:
       image_name = example[-1].item().split('.')[0]
-      # -1 because class labels are index-1 based.
-      # There are no labels in the "test" split.
-      label = -1 if split_name == 'test' else _NAMES[example[4].item() - 1]
+      if self.version < '2.0.0' and split_name == 'test':
+        label = -1
+      else:
+        label = _NAMES[example[4].item() - 1]
       image = image_dict[image_name]
       bbox = bbox_dict[image_name]
       yield image_name, {

tensorflow_datasets/image/cars196_test.py

@@ -20,11 +20,16 @@
 import tensorflow_datasets.testing as tfds_test
 
 
-class BinaryAlphaDigitsTest(tfds_test.DatasetBuilderTestCase):
+class Cars196Test(tfds_test.DatasetBuilderTestCase):
   DATASET_CLASS = cars196.Cars196
   SPLITS = {'train': 2, 'test': 2}
 
-  DL_EXTRACT_RESULT = {'train': 'train', 'test': 'test', 'extra': 'extra'}
+  DL_EXTRACT_RESULT = {
+      'train': 'train',
+      'test': 'test',
+      'extra': 'extra',
+      'test_annos': 'cars_test_annos_withlabels.mat'
+  }
 
 
 if __name__ == '__main__':

tensorflow_datasets/testing/test_data/fake_examples/cars196/cars_test_annos_withlabels.mat

@@ -1,3 +1,4 @@
 http://imagenet.stanford.edu/internal/car196/cars_test.tgz 977350468 bffea656d6f425cba3c91c6d83336e4c5f86c6cffd8975b0f375d3a10da8e243
+http://imagenet.stanford.edu/internal/car196/cars_test_annos_withlabels.mat 185758 790f75be8ea34eeded134cc559332baf23e30e91367e9ddca97d26ed9b895f05
 http://imagenet.stanford.edu/internal/car196/cars_train.tgz 979269282 512b227b30e2f0a8aab9e09485786ab4479582073a144998da74d64b801fd288
 https://ai.stanford.edu/~jkrause/cars/car_devkit.tgz 330960 b97deb463af7d58b6bfaa18b2a4de9829f0f79e8ce663dfa9261bf7810e9accd