def ConvolveUp(self, inputs, edge, target):
w = edge.params['weight']
conv = edge.conv_params
size = conv.size
stride = conv.stride
padding = conv.padding
num_filters = conv.num_filters
num_colors = conv.num_colors
f, numdims = w.shape
assert f == num_filters, 'f is %d but num_filters is %d' % (
f, num_filters)
assert numdims == size**2 * num_colors
input_t = edge.input_t
if conv.max_pool:
output_t = edge.unpooled_layer
else:
output_t = edge.output_t
numimages, numdims = input_t.shape
numimages2, numdims2 = output_t.shape
assert numimages == numimages2, '%d %d' % (numimages, numimages2)
assert numdims % num_colors == 0
x = int(np.sqrt(numdims / num_colors))
assert x**2 == numdims / num_colors
n_locs = (x + 2 * padding - size) / stride + 1
inputs.transpose(input_t)
cc.convUp(input_t, w, output_t, n_locs, padding, stride, num_colors)
# Do maxpooling
if conv.max_pool:
n_locs = (n_locs + 2 * padding -
conv.pool_size) / conv.pool_stride + 1
cc.MaxPool(output_t, edge.output_t, num_filters, conv.pool_size, 0,
conv.pool_stride, n_locs)
output_t = edge.output_t
output_t.transpose(target)
def ConvolveUp(self, inputs, edge, target):
w = edge.params['weight']
conv = edge.conv_params
size = conv.size
stride = conv.stride
padding = conv.padding
num_filters = conv.num_filters
num_colors = conv.num_colors
f, numdims = w.shape
assert f == num_filters, 'f is %d but num_filters is %d' % (f, num_filters)
assert numdims == size**2 * num_colors
input_t = edge.input_t
if conv.max_pool:
output_t = edge.unpooled_layer
else:
output_t = edge.output_t
numimages, numdims = input_t.shape
numimages2, numdims2 = output_t.shape
assert numimages == numimages2, '%d %d' % (numimages, numimages2)
assert numdims % num_colors == 0
x = int(np.sqrt(numdims / num_colors))
assert x**2 == numdims/num_colors
n_locs = (x + 2 * padding - size) / stride + 1
inputs.transpose(input_t)
cc.convUp(input_t, w, output_t, n_locs, padding, stride, num_colors)
# Do maxpooling
if conv.max_pool:
n_locs = (n_locs + 2 * padding - conv.pool_size) / conv.pool_stride + 1
cc.MaxPool(output_t, edge.output_t, num_filters, conv.pool_size, 0, conv.pool_stride, n_locs)
output_t = edge.output_t
output_t.transpose(target)
def ConvolveUp(self, inputs, edge, target):
w = edge.params["weight"]
conv = edge.conv_params
size = conv.size
stride = conv.stride
padding = conv.padding
num_filters = conv.num_filters
num_colors = conv.num_colors
f, numdims = w.shape
assert f == num_filters, "f is %d but num_filters is %d" % (f, num_filters)
if edge.conv:
assert numdims == size ** 2 * num_colors
input_t = edge.input_t
inputs.transpose(input_t)
# Convolve Up.
if conv.max_pool:
output_t = edge.unpooled_layer
elif conv.rnorm:
output_t = edge.unrnormalized_layer
else:
output_t = edge.output_t
numimages, numdims = input_t.shape
numimages2, numdims2 = output_t.shape
assert numimages == numimages2, "%d %d." % (numimages, numimages2)
assert numdims % num_colors == 0
x = int(np.sqrt(numdims / num_colors))
assert x ** 2 == numdims / num_colors
n_locs = (x + 2 * padding - size) / stride + 1
if edge.conv:
cc.convUp(input_t, w, output_t, n_locs, padding, stride, num_colors)
else:
cc.localUp(input_t, w, output_t, n_locs, padding, stride, num_colors)
# Do maxpooling
if conv.max_pool:
input_t = output_t
if conv.rnorm:
output_t = edge.unrnormalized_layer
else:
output_t = edge.output_t
n_locs = (n_locs - conv.pool_size) / conv.pool_stride + 1
if conv.prob:
# Get Gumbel variates.
rnd = edge.rnd
rnd.fill_with_rand()
cm.log(rnd)
rnd.mult(-1)
cm.log(rnd)
rnd.mult(-1)
cc.ProbMaxPool(input_t, rnd, output_t, num_filters, conv.pool_size, 0, conv.pool_stride, n_locs)
else:
cc.MaxPool(input_t, output_t, num_filters, conv.pool_size, 0, conv.pool_stride, n_locs)
if conv.rnorm:
input_t = output_t
output_t = edge.output_t
denoms = edge.denoms
sizeX = conv.norm_size
add_scale = conv.add_scale
pow_scale = conv.pow_scale
cc.ResponseNorm(input_t, denoms, output_t, num_filters, sizeX, add_scale, pow_scale)
output_t.transpose(target)
def ConvolveUp(inputs, edge, target):
w = edge.params['weight']
conv = edge.conv_params
size = conv.size
stride = conv.stride
padding = conv.padding
num_filters = conv.num_filters
num_colors = conv.num_colors
f, numdims = w.shape
assert f == num_filters, 'f is %d but num_filters is %d' % (f, num_filters)
if edge.conv:
assert numdims == size**2 * num_colors
input_t = edge.input_t
inputs.transpose(input_t)
# Convolve Up.
if conv.max_pool:
output_t = edge.unpooled_layer
elif conv.rnorm:
output_t = edge.unrnormalized_layer
else:
output_t = edge.output_t
numimages, numdims = input_t.shape
numimages2, numdims2 = output_t.shape
assert numimages == numimages2, '%d %d.' % (numimages, numimages2)
assert numdims % num_colors == 0
x = int(math.sqrt(numdims / num_colors))
assert x**2 == numdims/num_colors
n_locs = (x + 2 * padding - size) / stride + 1
if edge.conv:
cc.convUp(input_t, w, output_t, n_locs, padding, stride, num_colors)
else:
cc.localUp(input_t, w, output_t, n_locs, padding, stride, num_colors)
# Do maxpooling
if conv.max_pool:
input_t = output_t
if conv.rnorm:
output_t = edge.unrnormalized_layer
else:
output_t = edge.output_t
n_locs = (n_locs - conv.pool_size) / conv.pool_stride + 1
if conv.prob:
rnd = edge.rnd
rnd.fill_with_rand()
cm.log(rnd)
rnd.mult(-1)
#cm.log(rnd)
#rnd.mult(-1)
cc.ProbMaxPool(input_t, rnd, output_t, num_filters, conv.pool_size, 0, conv.pool_stride, n_locs)
else:
cc.MaxPool(input_t, output_t, num_filters, conv.pool_size, 0, conv.pool_stride, n_locs)
if conv.rnorm:
input_t = output_t
output_t = edge.output_t
denoms = edge.denoms
sizeX = conv.norm_size
add_scale = conv.add_scale
pow_scale = conv.pow_scale
cc.ResponseNorm(input_t, denoms, output_t, num_filters, sizeX, add_scale, pow_scale)
output_t.transpose(target)
