Add tests for VOCDetectionDataset

chainercv/datasets/voc/voc_detection_dataset.py

@@ -39,6 +39,8 @@ class VOCDetectionDataset(chainer.dataset.DatasetMixin):
 
     The array :obj:`difficult` is a one dimensional boolean array of shape
     :math:`(R,)`. :math:`R` is the number of bounding boxes in the image.
+    If :obj:`use_difficult` is :obj:`False`, this array is
+    a boolean array with all :obj:`False`.
 
     The type of the image, the bounding boxes and the labels are as follows.
 
@@ -76,7 +78,6 @@ def __init__(self, data_dir='auto', split='train', year='2012',
                     'for 2012 dataset. For 2007 dataset, you can pick \'test\''
                     ' in addition to the above mentioned splits.'
                 )
-
         id_list_file = os.path.join(
             data_dir, 'ImageSets/Main/{0}.txt'.format(split))
 
@@ -111,10 +112,10 @@ def get_example(self, i):
         for obj in anno.findall('object'):
             # when in not using difficult split, and the object is
             # difficult, skipt it.
-            difficult.append(int(obj.find('difficult').text))
             if not self.use_difficult and int(obj.find('difficult').text) == 1:
                 continue
 
+            difficult.append(int(obj.find('difficult').text))
             bndbox_anno = obj.find('bndbox')
             # subtract 1 to make pixel indexes 0-based
             bbox.append([
@@ -124,6 +125,7 @@ def get_example(self, i):
             label.append(voc_utils.voc_detection_label_names.index(name))
         bbox = np.stack(bbox).astype(np.float32)
         label = np.stack(label).astype(np.int32)
+        # When `use_difficult==False`, all elements in `difficult` are False.
         difficult = np.array(difficult, dtype=np.bool)
 
         # Load a image

tests/datasets_tests/test_voc/test_voc_detection_dataset.py

@@ -0,0 +1,79 @@
+import unittest
+
+import numpy as np
+
+from chainer import testing
+from chainer.testing import attr
+from chainer.testing import condition
+
+from chainercv.datasets import voc_detection_label_names
+from chainercv.datasets import VOCDetectionDataset
+
+
+def _create_paramters():
+    split_years = testing.product({
+        'split': ['train', 'trainval', 'val'],
+        'year': ['2007', '2012']})
+    split_years += [{'split': 'test', 'year': '2007'}]
+    params = testing.product_dict(
+        split_years,
+        [{'use_difficult': True, 'return_difficult': True},
+         {'use_difficult': True, 'return_difficult': False},
+         {'use_difficult': False, 'return_difficult': True},
+         {'use_difficult': False, 'return_difficult': False}])
+    return params
+
+
+@testing.parameterize(*_create_paramters())
+class TestVOCDetectionDataset(unittest.TestCase):
+
+    def setUp(self):
+        self.dataset = VOCDetectionDataset(
+            split=self.split,
+            year=self.year,
+            use_difficult=self.use_difficult,
+            return_difficult=self.return_difficult)
+        self.n_out = 4 if self.return_difficult else 3
+
+    @attr.slow
+    @condition.repeat(10)
+    def test_get_example(self):
+        i = np.random.randint(0, len(self.dataset))
+        out = self.dataset[i]
+
+        self.assertEqual(len(out), self.n_out)
+
+        img, bbox, label = out[:3]
+        C, H, W = img.shape
+
+        self.assertIsInstance(img, np.ndarray)
+        self.assertEqual(img.dtype, np.float32)
+        self.assertEqual(C, 3)
+        self.assertGreaterEqual(np.min(img), 0)
+        self.assertLessEqual(np.max(img), 255)
+
+        self.assertIsInstance(bbox, np.ndarray)
+        self.assertEqual(bbox.dtype, np.float32)
+        self.assertEqual(bbox.ndim, 2)
+        self.assertEqual(bbox.shape[1], 4)
+        np.testing.assert_array_less(bbox[:, 0], bbox[:, 2])
+        np.testing.assert_array_less(bbox[:, 1], bbox[:, 3])
+
+        self.assertIsInstance(label, np.ndarray)
+        self.assertEqual(label.dtype, np.int32)
+        self.assertEqual(label.shape, (bbox.shape[0],))
+        self.assertGreaterEqual(np.min(label), 0)
+        self.assertLessEqual(
+            np.max(label), len(voc_detection_label_names) - 1)
+
+        if self.n_out == 4:
+            difficult = out[3]
+            self.assertIsInstance(difficult, np.ndarray)
+            self.assertEqual(difficult.dtype, np.bool)
+            self.assertEqual(difficult.shape, (bbox.shape[0],))
+
+        if not self.use_difficult and self.return_difficult:
+            np.testing.assert_equal(difficult, 0)
+
+
+testing.run_module(__name__, __file__)