Skip to content
项目
群组
代码片段
帮助
当前项目
正在载入...
登录 / 注册
切换导航面板
P
pytensor
项目
项目
详情
活动
周期分析
仓库
仓库
文件
提交
分支
标签
贡献者
图表
比较
统计图
议题
0
议题
0
列表
看板
标记
里程碑
合并请求
0
合并请求
0
CI / CD
CI / CD
流水线
作业
日程
统计图
Wiki
Wiki
代码片段
代码片段
成员
成员
折叠边栏
关闭边栏
活动
图像
聊天
创建新问题
作业
提交
问题看板
Open sidebar
testgroup
pytensor
Commits
14766a3d
提交
14766a3d
authored
7月 07, 2017
作者:
João Victor Tozatti Risso
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
Refactor spatial transformer test to compare with CPU version
Signed-off-by:
João Victor Tozatti Risso
<
joaovictor.risso@gmail.com
>
上级
1cfebc92
显示空白字符变更
内嵌
并排
正在显示
1 个修改的文件
包含
166 行增加
和
77 行删除
+166
-77
test_dnn.py
theano/gpuarray/tests/test_dnn.py
+166
-77
没有找到文件。
theano/gpuarray/tests/test_dnn.py
浏览文件 @
14766a3d
...
@@ -2300,87 +2300,176 @@ class Cudnn_grouped_conv(Grouped_conv_noOptim):
...
@@ -2300,87 +2300,176 @@ class Cudnn_grouped_conv(Grouped_conv_noOptim):
is_dnn
=
True
is_dnn
=
True
def
test_dnn_spatialtf
_grid_generator
():
def
test_dnn_spatialtf
():
if
not
dnn
.
dnn_available
(
test_ctx_name
):
if
not
dnn
.
dnn_available
(
test_ctx_name
):
raise
SkipTest
(
dnn
.
dnn_available
.
msg
)
raise
SkipTest
(
dnn
.
dnn_available
.
msg
)
utt
.
seed_rng
()
utt
.
seed_rng
()
float_type
=
theano
.
config
.
floatX
"""
Spatial Transformer implementation using Theano from Lasagne
from
scipy
import
misc
Original author: skaae (https://github.com/skaae)
f
=
misc
.
face
(
gray
=
True
)
.
astype
(
float_type
)
"""
def
spatialtf_cpu
(
theta
,
inp
,
downsample_factor
,
border_mode
=
'nearest'
):
# shape: (num_images, channels, height, width), equivalent to NCHW
num_batch
,
num_channels
,
height
,
width
=
inp
.
shape
nchannels
=
f
.
shape
[
2
]
if
len
(
f
.
shape
)
==
3
else
1
theta
=
T
.
reshape
(
theta
,
(
-
1
,
2
,
3
))
assert
(
nchannels
is
not
None
)
grid_dims
=
(
3
,
3
,
128
,
128
)
# grid of (x_t, y_t, 1), eq (1) in ref [1]
out_height
=
T
.
cast
(
height
//
downsample_factor
,
'int64'
)
rotation
=
[[
-
1
,
0
,
0
],
out_width
=
T
.
cast
(
width
//
downsample_factor
,
'int64'
)
[
0
,
-
1
,
0
]]
grid
=
_meshgrid
(
out_height
,
out_width
)
# transform a x (x_t, y_t, 1)^t -> (x_s, y_s)
theta
=
np
.
asarray
(
grid_dims
[
0
]
*
[
rotation
],
dtype
=
float_type
)
t_g
=
T
.
dot
(
theta
,
grid
)
theta_gpu
=
gpuarray_shared_constructor
(
theta
)
x_s
=
t_g
[:,
0
]
y_s
=
t_g
[:,
1
]
def
normalize_input
(
input
):
x_s_flat
=
x_s
.
flatten
()
# Scale input from [0, 255] to [0, 2]
y_s_flat
=
y_s
.
flatten
()
scale_factor
=
2
**
-
7
# 1/128
input
*=
scale_factor
# dimshuffle input to (bs, height, width, channels)
# Re-scale input from [0, 2] to [-1, 1] (normalized)
input_dim
=
inp
.
dimshuffle
(
0
,
2
,
3
,
1
)
input
-=
1
input_transformed
=
_interpolate
(
return
input
input_dim
,
x_s_flat
,
y_s_flat
,
out_height
,
out_width
,
border_mode
)
def
rescale_input
(
input
):
# Re-scale output to range [0, 2]
output
=
T
.
reshape
(
input
+=
1
input_transformed
,
(
num_batch
,
out_height
,
out_width
,
num_channels
))
# Re-scale output to range [0, 255]
output
=
output
.
dimshuffle
(
0
,
3
,
1
,
2
)
# dimshuffle to conv format
input
*=
128
return
output
return
input
def
_interpolate
(
im
,
x
,
y
,
out_height
,
out_width
,
border_mode
):
# Gray-scale images don't have the channels dimension, only
# *_f are floats
# images with color channels have to converted from HWC to CHW
num_batch
,
height
,
width
,
channels
=
im
.
shape
if
len
(
f
.
shape
)
==
3
:
height_f
=
T
.
cast
(
height
,
theano
.
config
.
floatX
)
# Convert from HWC to CHW
width_f
=
T
.
cast
(
width
,
theano
.
config
.
floatX
)
f
=
np
.
transpose
(
f
,
axes
=
(
2
,
0
,
1
))
else
:
# scale coordinates from [-1, 1] to [0, width/height - 1]
# f.shape = (width, height)
x
=
(
x
+
1
)
/
2
*
(
width_f
-
1
)
shp
=
f
.
shape
y
=
(
y
+
1
)
/
2
*
(
height_f
-
1
)
# Add channel dimension
f
=
f
.
reshape
((
1
,
shp
[
0
],
shp
[
1
]))
# obtain indices of the 2x2 pixel neighborhood surrounding the coordinates;
# Normalize pixel values of the image in range [-1, 1]
# we need those in floatX for interpolation and in int64 for indexing.
f
=
normalize_input
(
f
)
x0_f
=
T
.
floor
(
x
)
# Create array of images
y0_f
=
T
.
floor
(
y
)
img
=
np
.
asarray
(
grid_dims
[
0
]
*
[
f
],
dtype
=
float_type
)
x1_f
=
x0_f
+
1
# Create GPU variable for the images
y1_f
=
y0_f
+
1
img_gpu
=
gpuarray_shared_constructor
(
img
)
# for indexing, we need to take care of the border mode for outside pixels.
spatialtf
=
dnn
.
dnn_spatialtf
(
img_gpu
,
theta_gpu
,
grid_dims
)
if
border_mode
==
'nearest'
:
x0
=
T
.
clip
(
x0_f
,
0
,
width_f
-
1
)
spatialtf_fn
=
theano
.
function
([],
[
spatialtf
],
mode
=
mode_with_gpu
)
x1
=
T
.
clip
(
x1_f
,
0
,
width_f
-
1
)
y0
=
T
.
clip
(
y0_f
,
0
,
height_f
-
1
)
result
,
=
spatialtf_fn
()
y1
=
T
.
clip
(
y1_f
,
0
,
height_f
-
1
)
elif
border_mode
==
'mirror'
:
img_out
=
np
.
asarray
(
result
,
dtype
=
float_type
)
w
=
2
*
(
width_f
-
1
)
# Re-scale image to range [0, 255]
x0
=
T
.
minimum
(
x0_f
%
w
,
-
x0_f
%
w
)
img_out
=
rescale_input
(
img_out
)
x1
=
T
.
minimum
(
x1_f
%
w
,
-
x1_f
%
w
)
# Convert to uint8 (byte)
h
=
2
*
(
height_f
-
1
)
img_out
=
img_out
.
astype
(
dtype
=
np
.
uint8
)
y0
=
T
.
minimum
(
y0_f
%
h
,
-
y0_f
%
h
)
# Transpose back to NHWC
y1
=
T
.
minimum
(
y1_f
%
h
,
-
y1_f
%
h
)
img_out
=
np
.
transpose
(
img_out
,
axes
=
(
0
,
2
,
3
,
1
))
elif
border_mode
==
'wrap'
:
x0
=
T
.
mod
(
x0_f
,
width_f
)
grayscale
=
False
x1
=
T
.
mod
(
x1_f
,
width_f
)
if
img_out
.
shape
[
3
]
==
1
:
# Gray-scale image
y0
=
T
.
mod
(
y0_f
,
height_f
)
grayscale
=
True
y1
=
T
.
mod
(
y1_f
,
height_f
)
shp
=
img_out
.
shape
img_out
=
img_out
.
reshape
(
shp
[
0
],
shp
[
1
],
shp
[
2
])
import
matplotlib.pyplot
as
plt
for
img_idx
in
range
(
len
(
img_out
)):
if
grayscale
:
plt
.
imshow
(
img_out
[
img_idx
],
cmap
=
'gray'
)
else
:
else
:
plt
.
imshow
(
img_out
[
img_idx
])
raise
ValueError
(
"border_mode must be one of "
plt
.
show
()
"'nearest', 'mirror', 'wrap'"
)
x0
,
x1
,
y0
,
y1
=
(
T
.
cast
(
v
,
'int64'
)
for
v
in
(
x0
,
x1
,
y0
,
y1
))
topo
=
spatialtf_fn
.
maker
.
fgraph
.
toposort
()
# The input is [num_batch, height, width, channels]. We do the lookup in
# the flattened input, i.e [num_batch*height*width, channels]. We need
# to offset all indices to match the flat version
dim2
=
width
dim1
=
width
*
height
base
=
T
.
repeat
(
T
.
arange
(
num_batch
,
dtype
=
'int64'
)
*
dim1
,
out_height
*
out_width
)
base_y0
=
base
+
y0
*
dim2
base_y1
=
base
+
y1
*
dim2
idx_a
=
base_y0
+
x0
idx_b
=
base_y1
+
x0
idx_c
=
base_y0
+
x1
idx_d
=
base_y1
+
x1
# use indices to lookup pixels for all samples
im_flat
=
im
.
reshape
((
-
1
,
channels
))
Ia
=
im_flat
[
idx_a
]
Ib
=
im_flat
[
idx_b
]
Ic
=
im_flat
[
idx_c
]
Id
=
im_flat
[
idx_d
]
# calculate interpolated values
wa
=
((
x1_f
-
x
)
*
(
y1_f
-
y
))
.
dimshuffle
(
0
,
'x'
)
wb
=
((
x1_f
-
x
)
*
(
y
-
y0_f
))
.
dimshuffle
(
0
,
'x'
)
wc
=
((
x
-
x0_f
)
*
(
y1_f
-
y
))
.
dimshuffle
(
0
,
'x'
)
wd
=
((
x
-
x0_f
)
*
(
y
-
y0_f
))
.
dimshuffle
(
0
,
'x'
)
output
=
T
.
sum
([
wa
*
Ia
,
wb
*
Ib
,
wc
*
Ic
,
wd
*
Id
],
axis
=
0
)
return
output
def
_linspace
(
start
,
stop
,
num
):
# Theano linspace. Behaves similar to np.linspace
start
=
T
.
cast
(
start
,
theano
.
config
.
floatX
)
stop
=
T
.
cast
(
stop
,
theano
.
config
.
floatX
)
num
=
T
.
cast
(
num
,
theano
.
config
.
floatX
)
step
=
(
stop
-
start
)
/
(
num
-
1
)
return
T
.
arange
(
num
,
dtype
=
theano
.
config
.
floatX
)
*
step
+
start
def
_meshgrid
(
height
,
width
):
# This function is the grid generator from eq. (1) in reference [1].
# It is equivalent to the following numpy code:
# x_t, y_t = np.meshgrid(np.linspace(-1, 1, width),
# np.linspace(-1, 1, height))
# ones = np.ones(np.prod(x_t.shape))
# grid = np.vstack([x_t.flatten(), y_t.flatten(), ones])
# It is implemented in Theano instead to support symbolic grid sizes.
# Note: If the image size is known at layer construction time, we could
# compute the meshgrid offline in numpy instead of doing it dynamically
# in Theano. However, it hardly affected performance when we tried.
x_t
=
T
.
dot
(
T
.
ones
((
height
,
1
)),
_linspace
(
-
1.0
,
1.0
,
width
)
.
dimshuffle
(
'x'
,
0
))
y_t
=
T
.
dot
(
_linspace
(
-
1.0
,
1.0
,
height
)
.
dimshuffle
(
0
,
'x'
),
T
.
ones
((
1
,
width
)))
x_t_flat
=
x_t
.
reshape
((
1
,
-
1
))
y_t_flat
=
y_t
.
reshape
((
1
,
-
1
))
ones
=
T
.
ones_like
(
x_t_flat
)
grid
=
T
.
concatenate
([
x_t_flat
,
y_t_flat
,
ones
],
axis
=
0
)
return
grid
# Generate random set of RGB images with 256x256 resolution (pixel values in [0, 255])
img_dims
=
(
10
,
256
,
256
,
3
)
# images are usually NHWC
img
=
np
.
random
.
randint
(
low
=
0
,
high
=
256
,
size
=
img_dims
)
# Convert from NHWC to NCHW
img
=
np
.
transpose
(
img
,
axes
=
(
0
,
3
,
1
,
2
))
.
astype
(
theano
.
config
.
floatX
)
gpu_img
=
gpuarray_shared_constructor
(
img
)
downsample_factor
=
2
grid_h
=
img_dims
[
1
]
//
downsample_factor
grid_w
=
img_dims
[
2
]
//
downsample_factor
grid_dims
=
(
img_dims
[
0
],
img_dims
[
3
],
grid_h
,
grid_w
)
# Transformation matrix
rotation
=
[[
1
,
0
,
0
],
[
0
,
1
,
0
]]
transform
=
np
.
asarray
(
img_dims
[
0
]
*
[
rotation
],
dtype
=
theano
.
config
.
floatX
)
gpu_transform
=
gpuarray_shared_constructor
(
transform
)
st_dnn
=
dnn
.
dnn_spatialtf
(
gpu_img
,
gpu_transform
,
grid_dims
)
st_dnn_func
=
theano
.
function
([],
[
st_dnn
])
# Check if function graph contains the spatial transformer Ops
topo
=
st_dnn_func
.
maker
.
fgraph
.
toposort
()
assert
len
([
n
for
n
in
topo
if
isinstance
(
n
.
op
,
dnn
.
GpuDnnGridGenerator
)])
==
1
assert
len
([
n
for
n
in
topo
if
isinstance
(
n
.
op
,
dnn
.
GpuDnnGridGenerator
)])
==
1
assert
len
([
n
for
n
in
topo
if
isinstance
(
n
.
op
,
dnn
.
GpuDnnGridSampler
)])
==
1
assert
len
([
n
for
n
in
topo
if
isinstance
(
n
.
op
,
dnn
.
GpuDnnGridSampler
)])
==
1
# Setup CPU Op
t_img
=
T
.
tensor4
(
'img'
)
t_theta
=
T
.
tensor3
(
'theta'
)
st_cpu
=
spatialtf_cpu
(
t_theta
,
t_img
,
downsample_factor
,
'nearest'
)
st_cpu_func
=
theano
.
function
([
t_theta
,
t_img
],
[
st_cpu
],
mode
=
mode_without_gpu
)
res
,
=
st_cpu_func
(
transform
,
img
)
img_out_gpu
=
st_dnn_func
()
img_out
=
np
.
asarray
(
img_out_gpu
[
0
])
utt
.
assert_allclose
(
img_out
,
res
,
rtol
=
1e-2
,
atol
=
1e-2
)
编写
预览
Markdown
格式
0%
重试
或
添加新文件
添加附件
取消
您添加了
0
人
到此讨论。请谨慎行事。
请先完成此评论的编辑!
取消
请
注册
或者
登录
后发表评论