Assert input type for transform functions

chainercv/experimental/links/model/pspnet/pspnet.py

@@ -72,16 +72,16 @@ def __init__(self, n_layer, initialW, bn_kwargs=None):
                 x, ksize=3, stride=2, pad=1)
             self.res2 = ResBlock(
                 n_block[0], 128, 64, 256, 1, 1,
-                initialW, bn_kwargs, stride_first=False)
+                initialW=initialW, bn_kwargs=bn_kwargs, stride_first=False)
             self.res3 = ResBlock(
                 n_block[1], 256, 128, 512, 2, 1,
-                initialW, bn_kwargs, stride_first=False)
+                initialW=initialW, bn_kwargs=bn_kwargs, stride_first=False)
             self.res4 = ResBlock(
                 n_block[2], 512, 256, 1024, 1, 2,
-                initialW, bn_kwargs, stride_first=False)
+                initialW=initialW, bn_kwargs=bn_kwargs, stride_first=False)
             self.res5 = ResBlock(
                 n_block[3], 1024, 512, 2048, 1, 4,
-                initialW, bn_kwargs, stride_first=False)
+                initialW=initialW, bn_kwargs=bn_kwargs, stride_first=False)
 
 
 class PSPNet(chainer.Chain):

chainercv/transforms/bbox/crop_bbox.py

@@ -1,5 +1,7 @@
 import numpy as np
 
+from chainercv.utils.testing.assertions.assert_is_bbox import assert_is_bbox
+
 
 def crop_bbox(
         bbox, y_slice=None, x_slice=None,
@@ -46,7 +48,7 @@ def crop_bbox(
             bounding boxes.
 
     """
-
+    assert_is_bbox(bbox)
     t, b = _slice_to_bounds(y_slice)
     l, r = _slice_to_bounds(x_slice)
     crop_bb = np.array((t, l, b, r))

chainercv/transforms/bbox/flip_bbox.py

@@ -1,3 +1,6 @@
+from chainercv.utils.testing.assertions.assert_is_bbox import assert_is_bbox
+
+
 def flip_bbox(bbox, size, y_flip=False, x_flip=False):
     """Flip bounding boxes accordingly.
 
@@ -23,6 +26,7 @@ def flip_bbox(bbox, size, y_flip=False, x_flip=False):
         Bounding boxes flipped according to the given flips.
 
     """
+    assert_is_bbox(bbox, size)
     H, W = size
     bbox = bbox.copy()
     if y_flip:

chainercv/transforms/bbox/resize_bbox.py

@@ -1,3 +1,6 @@
+from chainercv.utils.testing.assertions.assert_is_bbox import assert_is_bbox
+
+
 def resize_bbox(bbox, in_size, out_size):
     """Resize bounding boxes according to image resize.
 
@@ -21,6 +24,7 @@ def resize_bbox(bbox, in_size, out_size):
         Bounding boxes rescaled according to the given image shapes.
 
     """
+    assert_is_bbox(bbox)
     bbox = bbox.copy()
     y_scale = float(out_size[0]) / in_size[0]
     x_scale = float(out_size[1]) / in_size[1]

chainercv/transforms/bbox/translate_bbox.py

@@ -1,3 +1,6 @@
+from chainercv.utils.testing.assertions.assert_is_bbox import assert_is_bbox
+
+
 def translate_bbox(bbox, y_offset=0, x_offset=0):
     """Translate bounding boxes.
 
@@ -24,7 +27,7 @@ def translate_bbox(bbox, y_offset=0, x_offset=0):
         Bounding boxes translated according to the given offsets.
 
     """
-
+    assert_is_bbox(bbox)
     out_bbox = bbox.copy()
     out_bbox[:, :2] += (y_offset, x_offset)
     out_bbox[:, 2:] += (y_offset, x_offset)

chainercv/transforms/image/center_crop.py

@@ -1,3 +1,6 @@
+from chainercv.utils.testing.assertions.assert_is_image import assert_is_image
+
+
 def center_crop(img, size, return_param=False, copy=False):
     """Center crop an image by `size`.
 
@@ -35,6 +38,7 @@ def center_crop(img, size, return_param=False, copy=False):
                 out_img = img[:, y_slice, x_slice]
 
     """
+    assert_is_image(img, color=None)
     _, H, W = img.shape
     oH, oW = size
     if oH > H or oW > W:

chainercv/transforms/image/flip.py

@@ -1,3 +1,6 @@
+from chainercv.utils.testing.assertions.assert_is_image import assert_is_image
+
+
 def flip(img, y_flip=False, x_flip=False, copy=False):
     """Flip an image in vertical or horizontal direction as specified.
 
@@ -11,6 +14,8 @@ def flip(img, y_flip=False, x_flip=False, copy=False):
     Returns:
         Transformed :obj:`img` in CHW format.
     """
+
+    assert_is_image(img, color=None)
     if y_flip:
         img = img[:, ::-1, :]
     if x_flip:

chainercv/transforms/image/pca_lighting.py

@@ -1,5 +1,7 @@
 import numpy as np
 
+from chainercv.utils.testing.assertions.assert_is_image import assert_is_image
+
 
 def pca_lighting(img, sigma, eigen_value=None, eigen_vector=None):
     """AlexNet style color augmentation
@@ -30,6 +32,7 @@ def pca_lighting(img, sigma, eigen_value=None, eigen_vector=None):
         An image in CHW format.
     """
 
+    assert_is_image(img, color=True)
     if sigma <= 0:
         return img
 

chainercv/transforms/image/resize.py

@@ -2,6 +2,8 @@
 import PIL
 import warnings
 
+from chainercv.utils.testing.assertions.assert_is_image import assert_is_image
+
 
 try:
     import cv2
@@ -68,5 +70,6 @@ def resize(img, size, interpolation=PIL.Image.BILINEAR):
         ~numpy.ndarray: A resize array in CHW format.
 
     """
+    assert_is_image(img, color=None)
     img = _resize(img, size, interpolation)
     return img

chainercv/transforms/image/resize_contain.py

@@ -3,6 +3,7 @@
 import PIL
 
 from chainercv.transforms import resize
+from chainercv.utils.testing.assertions.assert_is_image import assert_is_image
 
 
 def resize_contain(img, size, fill=0, interpolation=PIL.Image.BILINEAR,
@@ -52,6 +53,7 @@ def resize_contain(img, size, fill=0, interpolation=PIL.Image.BILINEAR,
             :obj:`height, width`.
 
     """
+    assert_is_image(img, color=None)
     C, H, W = img.shape
     out_H, out_W = size
     scale_h = out_H / H

chainercv/transforms/image/scale.py

@@ -1,6 +1,7 @@
 import PIL
 
 from chainercv.transforms import resize
+from chainercv.utils.testing.assertions.assert_is_image import assert_is_image
 
 
 def scale(img, size, fit_short=True, interpolation=PIL.Image.BILINEAR):
@@ -30,6 +31,7 @@ def scale(img, size, fit_short=True, interpolation=PIL.Image.BILINEAR):
         ~numpy.ndarray: A scaled image in CHW format.
 
     """
+    assert_is_image(img, color=None)
     _, H, W = img.shape
 
     # If resizing is not necessary, return the input as is.

chainercv/transforms/image/ten_crop.py

@@ -1,5 +1,7 @@
 import numpy as np
 
+from chainercv.utils.testing.assertions.assert_is_image import assert_is_image
+
 
 def ten_crop(img, size):
     """Crop 10 regions from an array.
@@ -31,6 +33,7 @@ def ten_crop(img, size):
         The cropped arrays. The shape of tensor is :math:`(10, C, H, W)`.
 
     """
+    assert_is_image(img, color=None)
     H, W = size
     iH, iW = img.shape[1:3]
 

chainercv/utils/testing/assertions/assert_is_bbox.py

@@ -1,4 +1,4 @@
-import numpy as np
+from chainer.backends import cuda
 
 
 def assert_is_bbox(bbox, size=None):
@@ -16,9 +16,10 @@ def assert_is_bbox(bbox, size=None):
             Each bounding box should be within the image.
     """
 
-    assert isinstance(bbox, np.ndarray), \
-        'bbox must be a numpy.ndarray.'
-    assert bbox.dtype == np.float32, \
+    xp = cuda.get_array_module(bbox)
+    assert isinstance(bbox, xp.ndarray), \
+        'bbox must be a numpy.ndarray or cupy.ndarray.'
+    assert bbox.dtype == xp.float32, \
         'The type of bbox must be numpy.float32,'
     assert bbox.shape[1:] == (4,), \
         'The shape of bbox must be (*, 4).'

chainercv/utils/testing/assertions/assert_is_image.py

@@ -1,4 +1,4 @@
-import numpy as np
+from chainer.backends import cuda
 
 
 def assert_is_image(img, color=True, check_range=True):
@@ -22,13 +22,15 @@ def assert_is_image(img, color=True, check_range=True):
 
     """
 
-    assert isinstance(img, np.ndarray), 'img must be a numpy.ndarray.'
+    xp = cuda.get_array_module(img)
+    assert isinstance(img, xp.ndarray), \
+        'img must be a numpy.ndarray or cupy.ndarray.'
     assert len(img.shape) == 3, 'img must be a 3-dimensional array.'
     C, H, W = img.shape
 
     if color:
         assert C == 3, 'The number of channels must be 3.'
-    else:
+    elif color is False:
         assert C == 1, 'The number of channels must be 1.'
 
     if check_range:

tests/experimental_tests/links_tests/model_tests/fcis_tests/test_fcis.py

@@ -183,7 +183,8 @@ def setUp(self):
         )
 
     def check_prepare(self):
-        x = _random_array(np, self.in_shape)
+        x = np.random.randint(
+            0, 256, size=self.in_shape).astype(np.float32)
         out = self.link.prepare(x)
         self.assertIsInstance(out, np.ndarray)
         self.assertEqual(out.shape, self.expected_shape)

tests/links_tests/model_tests/faster_rcnn_tests/test_faster_rcnn.py

@@ -92,7 +92,8 @@ def setUp(self):
         )
 
     def check_prepare(self):
-        x = _random_array(np, self.in_shape)
+        x = np.random.randint(
+            0, 256, size=self.in_shape).astype(np.float32)
         out = self.link.prepare(x)
         self.assertIsInstance(out, np.ndarray)
         self.assertEqual(out.shape, self.expected_shape)

tests/transforms_tests/bbox_tests/test_crop_bbox.py

@@ -15,7 +15,7 @@ def setUp(self):
             (0, 5, 3, 6),
             (1, 2, 3, 4),
             (3, 3, 4, 6),
-        ))
+        ), dtype=np.float32)
         self.y_slice = slice(1, 5)
         self.x_slice = slice(0, 4)
 
@@ -25,7 +25,7 @@ def test_crop_bbox(self):
             (0, 0, 4, 4),
             (0, 2, 2, 4),
             (2, 3, 3, 4),
-        ))
+        ), dtype=np.float32)
 
         out, param = crop_bbox(
             self.bbox, y_slice=self.y_slice, x_slice=self.x_slice,
@@ -38,7 +38,7 @@ def test_crop_bbox_disallow_outside_center(self):
             (0, 0, 2, 4),
             (0, 0, 4, 4),
             (0, 2, 2, 4),
-        ))
+        ), dtype=np.float32)
 
         out, param = crop_bbox(
             self.bbox, y_slice=self.y_slice, x_slice=self.x_slice,

tests/transforms_tests/bbox_tests/test_flip_bbox.py

@@ -4,13 +4,13 @@
 
 from chainer import testing
 from chainercv.transforms import flip_bbox
+from chainercv.utils.testing.generate_random_bbox import generate_random_bbox
 
 
 class TestFlipBbox(unittest.TestCase):
 
     def test_flip_bbox(self):
-        bbox = np.random.uniform(
-            low=0., high=32., size=(10, 4))
+        bbox = generate_random_bbox(10, (32, 32), 0, 32)
 
         out = flip_bbox(bbox, size=(34, 32), y_flip=True)
         bbox_expected = bbox.copy()

tests/transforms_tests/bbox_tests/test_resize_bbox.py

@@ -4,13 +4,13 @@
 
 from chainer import testing
 from chainercv.transforms import resize_bbox
+from chainercv.utils.testing.generate_random_bbox import generate_random_bbox
 
 
 class TestResizeBbox(unittest.TestCase):
 
     def test_resize_bbox(self):
-        bbox = np.random.uniform(
-            low=0., high=32., size=(10, 4))
+        bbox = generate_random_bbox(10, (32, 32), 0, 32)
 
         out = resize_bbox(bbox, in_size=(32, 32), out_size=(64, 128))
         bbox_expected = bbox.copy()

tests/transforms_tests/bbox_tests/test_translate_bbox.py

@@ -4,13 +4,13 @@
 
 from chainer import testing
 from chainercv.transforms import translate_bbox
+from chainercv.utils.testing.generate_random_bbox import generate_random_bbox
 
 
 class TestTranslateBbox(unittest.TestCase):
 
     def test_translate_bbox(self):
-        bbox = np.random.uniform(
-            low=0., high=32., size=(10, 4))
+        bbox = generate_random_bbox(10, (32, 32), 0, 32)
 
         out = translate_bbox(bbox, y_offset=5, x_offset=3)
         bbox_expected = np.empty_like(bbox)