def __init__(self, **links):
super(VGG16, self).__init__(
conv1_1=L.Convolution2D(None, 64, 3, pad=1),
conv1_2=L.Convolution2D(None, 64, 3, pad=1),
conv2_1=L.Convolution2D(None, 128, 3, pad=1),
conv2_2=L.Convolution2D(None, 128, 3, pad=1),
conv3_1=L.Convolution2D(None, 256, 3, pad=1),
conv3_2=L.Convolution2D(None, 256, 3, pad=1),
conv3_3=L.Convolution2D(None, 256, 3, pad=1),
conv4_1=L.Convolution2D(None, 512, 3, pad=1),
conv4_2=L.Convolution2D(None, 512, 3, pad=1),
conv4_3=L.Convolution2D(None, 512, 3, pad=1),
norm4=Normalize(512, initial=initializers.Constant(20)),
conv5_1=L.DilatedConvolution2D(None, 512, 3, pad=1),
conv5_2=L.DilatedConvolution2D(None, 512, 3, pad=1),
conv5_3=L.DilatedConvolution2D(None, 512, 3, pad=1),
conv6=L.DilatedConvolution2D(None, 1024, 3, pad=6, dilate=6),
conv7=L.Convolution2D(None, 1024, 1),
)
for name, link in six.iteritems(links):
self.add_link(name, link)
class TestNormalize(unittest.TestCase):
def setUp(self):
self.link = Normalize(
self.n_channel, initializers.Normal(), eps=self.eps)
self.x = np.random.uniform(size=(1, self.n_channel) + self.shape) \
.astype(np.float32)
def _check_forward(self, x):
y = self.link(x)
self.assertIsInstance(y, chainer.Variable)
self.assertIsInstance(y.array, type(x))
self.assertEqual(y.shape, x.shape)
self.assertEqual(y.dtype, x.dtype)
x = chainer.backends.cuda.to_cpu(x)
y = chainer.backends.cuda.to_cpu(y.array)
scale = chainer.backends.cuda.to_cpu(self.link.scale.array)
norm = np.linalg.norm(x, axis=1, keepdims=True) + self.eps
expect = x / norm * scale[:, np.newaxis, np.newaxis]
np.testing.assert_almost_equal(y, expect)
def test_forward_cpu(self):
self._check_forward(self.x)
@attr.gpu
def test_forward_gpu(self):
self.link.to_gpu()
self._check_forward(chainer.backends.cuda.to_gpu(self.x))
def test_forward_zero_cpu(self):
self._check_forward(np.zeros_like(self.x))
@attr.gpu
def test_forward_zero_gpu(self):
self.link.to_gpu()
self._check_forward(
chainer.backends.cuda.to_gpu(np.zeros_like(self.x)))
class TestNormalize(unittest.TestCase):
def setUp(self):
self.link = Normalize(
self.n_channel, initializers.Normal(), eps=self.eps)
self.x = np.random.uniform(size=(1, self.n_channel) + self.shape) \
.astype(np.float32)
def _check_forward(self, x):
y = self.link(x)
self.assertIsInstance(y, chainer.Variable)
self.assertIsInstance(y.array, type(x))
self.assertEqual(y.shape, x.shape)
self.assertEqual(y.dtype, x.dtype)
x = chainer.cuda.to_cpu(x)
y = chainer.cuda.to_cpu(y.array)
scale = chainer.cuda.to_cpu(self.link.scale.array)
norm = np.linalg.norm(x, axis=1, keepdims=True) + self.eps
expect = x / norm * scale[:, np.newaxis, np.newaxis]
np.testing.assert_almost_equal(y, expect)
def test_forward_cpu(self):
self._check_forward(self.x)
@attr.gpu
def test_forward_gpu(self):
self.link.to_gpu()
self._check_forward(chainer.cuda.to_gpu(self.x))
def test_forward_zero_cpu(self):
self._check_forward(np.zeros_like(self.x))
@attr.gpu
def test_forward_zero_gpu(self):
self.link.to_gpu()
self._check_forward(chainer.cuda.to_gpu(np.zeros_like(self.x)))
def __init__(self):
super(VGG16, self).__init__()
with self.init_scope():
self.conv1_1 = L.Convolution2D(64, 3, pad=1)
self.conv1_2 = L.Convolution2D(64, 3, pad=1)
self.conv2_1 = L.Convolution2D(128, 3, pad=1)
self.conv2_2 = L.Convolution2D(128, 3, pad=1)
self.conv3_1 = L.Convolution2D(256, 3, pad=1)
self.conv3_2 = L.Convolution2D(256, 3, pad=1)
self.conv3_3 = L.Convolution2D(256, 3, pad=1)
self.conv4_1 = L.Convolution2D(512, 3, pad=1)
self.conv4_2 = L.Convolution2D(512, 3, pad=1)
self.conv4_3 = L.Convolution2D(512, 3, pad=1)
self.norm4 = Normalize(512, initial=initializers.Constant(20))
self.conv5_1 = L.DilatedConvolution2D(512, 3, pad=1)
self.conv5_2 = L.DilatedConvolution2D(512, 3, pad=1)
self.conv5_3 = L.DilatedConvolution2D(512, 3, pad=1)
self.conv6 = L.DilatedConvolution2D(1024, 3, pad=6, dilate=6)
self.conv7 = L.Convolution2D(1024, 1)
def _setup_normalize(self, layer):
blobs = layer.blobs
func = Normalize(caffe._get_num(blobs[0]))
func.scale.array[:] = np.array(blobs[0].data)
with self.init_scope():
setattr(self, layer.name, func)
def setUp(self):
self.link = Normalize(
self.n_channel, initializers.Normal(), eps=self.eps)
self.x = np.random.uniform(size=(1, self.n_channel) + self.shape) \
.astype(np.float32)
def setUp(self):
self.link = Normalize(
self.n_channel, initializers.Normal(), eps=self.eps)
self.x = np.random.uniform(size=(1, self.n_channel) + self.shape) \
.astype(np.float32)
def _setup_normarize(self, layer):
blobs = layer.blobs
func = Normalize(caffe._get_num(blobs[0]))
func.scale.data[:] = np.array(blobs[0].data)
self.add_link(layer.name, func)