def setUp(self):
# Spacial pyramid pooling uses max pooling in its implementation.
# To avoid instability of numerical gradient, use enough different
# values.
shape = (self.n, self.c, self.h, self.w)
self.x = pooling_nd_helper.shuffled_linspace(shape, self.dtype)
self.one = numpy.ones(
(self.n, self.c, self.h, self.w)).astype(self.dtype)
self.gy = numpy.random.uniform(
-1, 1, (self.n, self.output_dim, 1, 1)).astype(self.dtype)
self.ggx = numpy.random.uniform(-1, 1, shape).astype(self.dtype)
def setUp(self):
self.N = 2
self.n_channels = 3
inh, inw = self.insize
self.x = pooling_nd_helper.shuffled_linspace(
(self.N, self.n_channels, inh, inw), self.dtype)
outh, outw = self.outsize or self.expected_outsize
self.gy = numpy.random.uniform(
-1, 1, (self.N, self.n_channels, outh, outw)).astype(self.dtype)
self.check_backward_options = {'atol': 1e-4, 'rtol': 1e-3}
self.check_double_backward_options = {}
if self.dtype == numpy.float16:
self.check_backward_options = {'atol': 2e-3, 'rtol': 2e-2}
self.check_double_backward_options = {'atol': 3e-3, 'rtol': 3e-2}
self.ggx = numpy.random.uniform(-1, 1, self.x.shape).astype(self.dtype)
def setUp(self):
N = 3
n_channels = 3
self.x = pooling_nd_helper.shuffled_linspace(
(N, n_channels, 12, 8), numpy.float32)
self.rois = numpy.array([
[1, 1, 6, 6],
[6, 2, 7, 11],
[3, 1, 5, 10],
[3, 3, 3, 3],
[1.1, 2.2, 3.3, 4.4],
], dtype=numpy.float32)
self.roi_indices = numpy.array([0, 2, 1, 0, 2], dtype=numpy.int32)
n_rois = self.rois.shape[0]
outsize = _pair(self.outsize)
self.gy = numpy.random.uniform(
-1, 1, (n_rois, n_channels,
outsize[0], outsize[1])).astype(numpy.float32)
self.check_backward_options = {'atol': 5e-4, 'rtol': 5e-3}
def setUp(self):
N = 3
n_channels = 3
self.x = pooling_nd_helper.shuffled_linspace(
(N, n_channels, 12, 8), self.dtype)
self.rois = numpy.array([
[1, 1, 7, 7],
[2, 6, 12, 8],
[1, 3, 11, 6],
[3, 3, 4, 4]
], dtype=self.dtype)
self.roi_indices = numpy.array([0, 2, 1, 0], dtype=numpy.int32)
n_rois = self.rois.shape[0]
outsize = _pair(self.outsize)
self.gy = numpy.random.uniform(
-1, 1, (n_rois, n_channels,
outsize[0], outsize[1])).astype(self.dtype)
if self.dtype == numpy.float16:
self.check_backward_options = {
'dtype': numpy.float64, 'atol': 1e-2, 'rtol': 1e-2}
else:
self.check_backward_options = {'atol': 1e-3, 'rtol': 1e-2}
def setUp(self):
N = 3
n_channels = 3
self.x = pooling_nd_helper.shuffled_linspace((N, n_channels, 12, 8),
self.dtype)
self.rois = numpy.array([[0, 1, 1, 6, 6], [2, 6, 2, 7, 11],
[1, 3, 1, 5, 10], [0, 3, 3, 3, 3]],
dtype=self.dtype)
n_rois = self.rois.shape[0]
self.outh, self.outw = 5, 7
self.spatial_scale = 0.6
self.gy = numpy.random.uniform(
-1, 1,
(n_rois, n_channels, self.outh, self.outw)).astype(self.dtype)
if self.dtype == numpy.float16:
self.check_backward_options = {
'dtype': numpy.float64,
'atol': 1e-2,
'rtol': 1e-2
}
else:
self.check_backward_options = {'atol': 1e-3, 'rtol': 1e-2}
def generate_inputs(self):
return pooling_nd_helper.shuffled_linspace((2, 3, 4, 3), self.dtype),
def setUp(self):
self.x = pooling_nd_helper.shuffled_linspace((2, 3, 4, 4),
numpy.float32)