You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@mxnet.apache.org by GitBox <gi...@apache.org> on 2017/12/16 02:28:01 UTC
[GitHub] dwSun closed pull request #8894: Mobilenet
dwSun closed pull request #8894: Mobilenet
URL: https://github.com/apache/incubator-mxnet/pull/8894
This is a PR merged from a forked repository.
As GitHub hides the original diff on merge, it is displayed below for
the sake of provenance:
As this is a foreign pull request (from a fork), the diff is supplied
below (as it won't show otherwise due to GitHub magic):
diff --git a/example/image-classification/symbols/mobilenet.py b/example/image-classification/symbols/mobilenet.py
index 42b9636261..64b6e40fbf 100644
--- a/example/image-classification/symbols/mobilenet.py
+++ b/example/image-classification/symbols/mobilenet.py
@@ -14,48 +14,129 @@
# KIND, either express or implied. See the License for the
# specific language governing permissions and limitations
# under the License.
-
+# -*- coding:utf-8 -*-
import mxnet as mx
-def Conv(data, num_filter=1, kernel=(1, 1), stride=(1, 1), pad=(0, 0), num_group=1, name=None, suffix=''):
- conv = mx.sym.Convolution(data=data, num_filter=num_filter, kernel=kernel, num_group=num_group, stride=stride, pad=pad, no_bias=True, name='%s%s_conv2d' %(name, suffix))
- bn = mx.sym.BatchNorm(data=conv, name='%s%s_batchnorm' %(name, suffix), fix_gamma=True)
- act = mx.sym.Activation(data=bn, act_type='relu', name='%s%s_relu' %(name, suffix))
+__author__ = 'qingzhouzhen'
+modified_date = '17/8/5'
+__modify__ = 'dwSun'
+modified_date = '17/11/30'
+
+'''
+mobilenet
+ Suittable for image with around resolution x resolution, resolution is multiple of 32.
+
+Reference:
+ MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications
+ https://arxiv.org/abs/1704.04861
+'''
+
+alpha_values = [0.25, 0.50, 0.75, 1.0]
+
+
+def Conv(data, num_filter=1, kernel=(1, 1), stride=(1, 1), pad=(0, 0), num_group=1, name='', suffix=''):
+ conv = mx.sym.Convolution(data=data, num_filter=num_filter, kernel=kernel, num_group=num_group, stride=stride, pad=pad, no_bias=True, name='%s%s_conv2d' % (name, suffix))
+ bn = mx.sym.BatchNorm(data=conv, name='%s%s_batchnorm' % (name, suffix), fix_gamma=True)
+ act = mx.sym.Activation(data=bn, act_type='relu', name='%s%s_relu' % (name, suffix))
return act
-def get_symbol(num_classes, **kwargs):
- data = mx.symbol.Variable(name="data") # 224
- conv_1 = Conv(data, num_filter=32, kernel=(3, 3), pad=(1, 1), stride=(2, 2), name="conv_1") # 224/112
- conv_2_dw = Conv(conv_1, num_group=32, num_filter=32, kernel=(3, 3), pad=(1, 1), stride=(1, 1), name="conv_2_dw") # 112/112
- conv_2 = Conv(conv_2_dw, num_filter=64, kernel=(1, 1), pad=(0, 0), stride=(1, 1), name="conv_2") # 112/112
- conv_3_dw = Conv(conv_2, num_group=64, num_filter=64, kernel=(3, 3), pad=(1, 1), stride=(2, 2), name="conv_3_dw") # 112/56
- conv_3 = Conv(conv_3_dw, num_filter=128, kernel=(1, 1), pad=(0, 0), stride=(1, 1), name="conv_3") # 56/56
- conv_4_dw = Conv(conv_3, num_group=128, num_filter=128, kernel=(3, 3), pad=(1, 1), stride=(1, 1), name="conv_4_dw") # 56/56
- conv_4 = Conv(conv_4_dw, num_filter=128, kernel=(1, 1), pad=(0, 0), stride=(1, 1), name="conv_4") # 56/56
- conv_5_dw = Conv(conv_4, num_group=128, num_filter=128, kernel=(3, 3), pad=(1, 1), stride=(2, 2), name="conv_5_dw") # 56/28
- conv_5 = Conv(conv_5_dw, num_filter=256, kernel=(1, 1), pad=(0, 0), stride=(1, 1), name="conv_5") # 28/28
- conv_6_dw = Conv(conv_5, num_group=256, num_filter=256, kernel=(3, 3), pad=(1, 1), stride=(1, 1), name="conv_6_dw") # 28/28
- conv_6 = Conv(conv_6_dw, num_filter=256, kernel=(1, 1), pad=(0, 0), stride=(1, 1), name="conv_6") # 28/28
- conv_7_dw = Conv(conv_6, num_group=256, num_filter=256, kernel=(3, 3), pad=(1, 1), stride=(2, 2), name="conv_7_dw") # 28/14
- conv_7 = Conv(conv_7_dw, num_filter=512, kernel=(1, 1), pad=(0, 0), stride=(1, 1), name="conv_7") # 14/14
-
- conv_8_dw = Conv(conv_7, num_group=512, num_filter=512, kernel=(3, 3), pad=(1, 1), stride=(1, 1), name="conv_8_dw") # 14/14
- conv_8 = Conv(conv_8_dw, num_filter=512, kernel=(1, 1), pad=(0, 0), stride=(1, 1), name="conv_8") # 14/14
- conv_9_dw = Conv(conv_8, num_group=512, num_filter=512, kernel=(3, 3), pad=(1, 1), stride=(1, 1), name="conv_9_dw") # 14/14
- conv_9 = Conv(conv_9_dw, num_filter=512, kernel=(1, 1), pad=(0, 0), stride=(1, 1), name="conv_9") # 14/14
- conv_10_dw = Conv(conv_9, num_group=512, num_filter=512, kernel=(3, 3), pad=(1, 1), stride=(1, 1), name="conv_10_dw") # 14/14
- conv_10 = Conv(conv_10_dw, num_filter=512, kernel=(1, 1), pad=(0, 0), stride=(1, 1), name="conv_10") # 14/14
- conv_11_dw = Conv(conv_10, num_group=512, num_filter=512, kernel=(3, 3), pad=(1, 1), stride=(1, 1), name="conv_11_dw") # 14/14
- conv_11 = Conv(conv_11_dw, num_filter=512, kernel=(1, 1), pad=(0, 0), stride=(1, 1), name="conv_11") # 14/14
- conv_12_dw = Conv(conv_11, num_group=512, num_filter=512, kernel=(3, 3), pad=(1, 1), stride=(1, 1), name="conv_12_dw") # 14/14
- conv_12 = Conv(conv_12_dw, num_filter=512, kernel=(1, 1), pad=(0, 0), stride=(1, 1), name="conv_12") # 14/14
-
- conv_13_dw = Conv(conv_12, num_group=512, num_filter=512, kernel=(3, 3), pad=(1, 1), stride=(2, 2), name="conv_13_dw") # 14/7
- conv_13 = Conv(conv_13_dw, num_filter=1024, kernel=(1, 1), pad=(0, 0), stride=(1, 1), name="conv_13") # 7/7
- conv_14_dw = Conv(conv_13, num_group=1024, num_filter=1024, kernel=(3, 3), pad=(1, 1), stride=(1, 1), name="conv_14_dw") # 7/7
- conv_14 = Conv(conv_14_dw, num_filter=1024, kernel=(1, 1), pad=(0, 0), stride=(1, 1), name="conv_14") # 7/7
-
- pool = mx.sym.Pooling(data=conv_14, kernel=(7, 7), stride=(1, 1), pool_type="avg", name="global_pool")
+
+def Conv_DPW(data, depth=1, stride=(1, 1), name='', idx=0, suffix=''):
+ conv_dw = Conv(data, num_group=depth, num_filter=depth, kernel=(3, 3), pad=(1, 1), stride=stride, name="conv_%d_dw" % (idx), suffix=suffix)
+ conv = Conv(conv_dw, num_filter=depth * stride[0], kernel=(1, 1), pad=(0, 0), stride=(1, 1), name="conv_%d" % (idx), suffix=suffix)
+ return conv
+
+
+def get_symbol_compact(num_classes, alpha=1, resolution=224, **kwargs):
+ assert alpha in alpha_values, 'Invalid alpha={0}, must be one of {1}'.format(alpha, alpha_values)
+ assert resolution % 32 == 0, 'resolution must be multiple of 32'
+
+ base = int(32 * alpha)
+
+ data = mx.symbol.Variable(name="data") # 224
+ conv_1 = Conv(data, num_filter=base, kernel=(3, 3), pad=(1, 1), stride=(2, 2), name="conv_1") # 32*alpha, 224/112
+
+ conv_2_dw = Conv(conv_1, num_group=base, num_filter=base, kernel=(3, 3), pad=(1, 1), stride=(1, 1), name="conv_2_dw") # 112/112
+ conv_2 = Conv(conv_2_dw, num_filter=base * 2, kernel=(1, 1), pad=(0, 0), stride=(1, 1), name="conv_2") # 32*alpha, 112/112
+
+ conv_3_dpw = Conv_DPW(conv_2, depth=base * 2, stride=(2, 2), idx=3) # 64*alpha, 112/56 => 56/56
+ conv_4_dpw = Conv_DPW(conv_3_dpw, depth=base * 4, stride=(1, 1), idx=4) # 128*alpha, 56/56 =>56/56
+ conv_5_dpw = Conv_DPW(conv_4_dpw, depth=base * 4, stride=(2, 2), idx=5) # 128*alpha, 56/28 => 28/28
+ conv_6_dpw = Conv_DPW(conv_5_dpw, depth=base * 8, stride=(1, 1), idx=6) # 256*alpha, 28/28 => 28/28
+ conv_7_dpw = Conv_DPW(conv_6_dpw, depth=base * 8, stride=(2, 2), idx=7) # 256*alpha, 28/14 => 14/14
+ conv_dpw = conv_7_dpw
+
+ for idx in range(8, 13):
+ conv_dpw = Conv_DPW(conv_dpw, depth=base * 16, stride=(1, 1), idx=idx) # 512*alpha, 14/14
+
+ conv_12_dpw = conv_dpw
+ conv_13_dpw = Conv_DPW(conv_12_dpw, depth=base * 16, stride=(2, 2), idx=13) # 512*alpha, 14/7 => 7/7
+ conv_14_dpw = Conv_DPW(conv_13_dpw, depth=base * 32, stride=(1, 1), idx=14) # 1024*alpha, 7/7 => 7/7
+
+ pool_size = int(resolution / 32)
+ pool = mx.sym.Pooling(data=conv_14_dpw, kernel=(pool_size, pool_size), stride=(1, 1), pool_type="avg", name="global_pool")
+ flatten = mx.sym.Flatten(data=pool, name="flatten")
+ fc = mx.symbol.FullyConnected(data=flatten, num_hidden=num_classes, name='fc')
+ softmax = mx.symbol.SoftmaxOutput(data=fc, name='softmax')
+ return softmax
+
+
+def get_symbol(num_classes, alpha=1, resolution=224, **kwargs):
+ assert alpha in alpha_values, 'Invalid alpha=[{0}], must be one of [{1}]'.format(alpha, alpha_values)
+ assert resolution % 32 == 0, 'resolution must be multpile of 32'
+
+ base = int(32 * alpha)
+
+ data = mx.symbol.Variable(name="data") # 224
+ depth = base # 32*alpha
+ conv_1 = Conv(data, num_filter=depth, kernel=(3, 3), pad=(1, 1), stride=(2, 2), name="conv_1") # 224/112
+
+ depth = base # 32*alpha
+ conv_2_dw = Conv(conv_1, num_group=depth, num_filter=depth, kernel=(3, 3), pad=(1, 1), stride=(1, 1), name="conv_2_dw") # 112/112
+ conv_2 = Conv(conv_2_dw, num_filter=depth * 2, kernel=(1, 1), pad=(0, 0), stride=(1, 1), name="conv_2") # 112/112
+
+ depth = base * 2 # 64*alpha
+ conv_3_dw = Conv(conv_2, num_group=depth, num_filter=depth, kernel=(3, 3), pad=(1, 1), stride=(2, 2), name="conv_3_dw") # 112/56
+ conv_3 = Conv(conv_3_dw, num_filter=depth * 2, kernel=(1, 1), pad=(0, 0), stride=(1, 1), name="conv_3") # 56/56
+
+ depth = base * 4 # 128*alpha
+ conv_4_dw = Conv(conv_3, num_group=depth, num_filter=depth, kernel=(3, 3), pad=(1, 1), stride=(1, 1), name="conv_4_dw") # 56/56
+ conv_4 = Conv(conv_4_dw, num_filter=depth, kernel=(1, 1), pad=(0, 0), stride=(1, 1), name="conv_4") # 56/56
+
+ depth = base * 4 # 128*alpha
+ conv_5_dw = Conv(conv_4, num_group=depth, num_filter=depth, kernel=(3, 3), pad=(1, 1), stride=(2, 2), name="conv_5_dw") # 56/28
+ conv_5 = Conv(conv_5_dw, num_filter=depth * 2, kernel=(1, 1), pad=(0, 0), stride=(1, 1), name="conv_5") # 28/28
+
+ depth = base * 8 # 256*alpha
+ conv_6_dw = Conv(conv_5, num_group=depth, num_filter=depth, kernel=(3, 3), pad=(1, 1), stride=(1, 1), name="conv_6_dw") # 28/28
+ conv_6 = Conv(conv_6_dw, num_filter=depth, kernel=(1, 1), pad=(0, 0), stride=(1, 1), name="conv_6") # 28/28
+
+ depth = base * 8 # 256*alpha
+ conv_7_dw = Conv(conv_6, num_group=depth, num_filter=depth, kernel=(3, 3), pad=(1, 1), stride=(2, 2), name="conv_7_dw") # 28/14
+ conv_7 = Conv(conv_7_dw, num_filter=depth * 2, kernel=(1, 1), pad=(0, 0), stride=(1, 1), name="conv_7") # 14/14
+
+ depth = base * 16 # 512*alpha
+ conv_8_dw = Conv(conv_7, num_group=depth, num_filter=depth, kernel=(3, 3), pad=(1, 1), stride=(1, 1), name="conv_8_dw") # 14/14
+ conv_8 = Conv(conv_8_dw, num_filter=depth, kernel=(1, 1), pad=(0, 0), stride=(1, 1), name="conv_8") # 14/14
+ conv_9_dw = Conv(conv_8, num_group=depth, num_filter=depth, kernel=(3, 3), pad=(1, 1), stride=(1, 1), name="conv_9_dw") # 14/14
+ conv_9 = Conv(conv_9_dw, num_filter=depth, kernel=(1, 1), pad=(0, 0), stride=(1, 1), name="conv_9") # 14/14
+ conv_10_dw = Conv(conv_9, num_group=depth, num_filter=depth, kernel=(3, 3), pad=(1, 1), stride=(1, 1), name="conv_10_dw") # 14/14
+ conv_10 = Conv(conv_10_dw, num_filter=depth, kernel=(1, 1), pad=(0, 0), stride=(1, 1), name="conv_10") # 14/14
+ conv_11_dw = Conv(conv_10, num_group=depth, num_filter=depth, kernel=(3, 3), pad=(1, 1), stride=(1, 1), name="conv_11_dw") # 14/14
+ conv_11 = Conv(conv_11_dw, num_filter=depth, kernel=(1, 1), pad=(0, 0), stride=(1, 1), name="conv_11") # 14/14
+ conv_12_dw = Conv(conv_11, num_group=depth, num_filter=depth, kernel=(3, 3), pad=(1, 1), stride=(1, 1), name="conv_12_dw") # 14/14
+ conv_12 = Conv(conv_12_dw, num_filter=depth, kernel=(1, 1), pad=(0, 0), stride=(1, 1), name="conv_12") # 14/14
+
+ depth = base * 16 # 512*alpha
+ conv_13_dw = Conv(conv_12, num_group=depth, num_filter=depth, kernel=(3, 3), pad=(1, 1), stride=(2, 2), name="conv_13_dw") # 14/7
+ conv_13 = Conv(conv_13_dw, num_filter=depth * 2, kernel=(1, 1), pad=(0, 0), stride=(1, 1), name="conv_13") # 7/7
+
+ depth = base * 32 # 1024*alpha
+ conv_14_dw = Conv(conv_13, num_group=depth, num_filter=depth, kernel=(3, 3), pad=(1, 1), stride=(1, 1), name="conv_14_dw") # 7/7
+ conv_14 = Conv(conv_14_dw, num_filter=depth, kernel=(1, 1), pad=(0, 0), stride=(1, 1), name="conv_14") # 7/7
+
+ pool_size = int(resolution / 32)
+ pool = mx.sym.Pooling(data=conv_14, kernel=(pool_size, pool_size), stride=(1, 1), pool_type="avg", name="global_pool")
flatten = mx.sym.Flatten(data=pool, name="flatten")
fc = mx.symbol.FullyConnected(data=flatten, num_hidden=num_classes, name='fc')
softmax = mx.symbol.SoftmaxOutput(data=fc, name='softmax')
diff --git a/tools/im2rec.py b/tools/im2rec.py
index ec6de19694..b2acdfaadb 100644
--- a/tools/im2rec.py
+++ b/tools/im2rec.py
@@ -35,6 +35,7 @@
except ImportError:
multiprocessing = None
+
def list_image(root, recursive, exts):
i = 0
if recursive:
@@ -50,8 +51,12 @@ def list_image(root, recursive, exts):
cat[path] = len(cat)
yield (i, os.path.relpath(fpath, root), cat[path])
i += 1
- for k, v in sorted(cat.items(), key=lambda x: x[1]):
- print(os.path.relpath(k, root), v)
+
+ with open(args.prefix + '_synsets.txt', 'w') as fout:
+ for k, v in sorted(cat.items(), key=lambda x: x[1]):
+ print(os.path.relpath(k, root), v)
+ fout.write('%s\n' % (os.path.relpath(k, root)))
+
else:
for fname in sorted(os.listdir(root)):
fpath = os.path.join(root, fname)
@@ -60,6 +65,7 @@ def list_image(root, recursive, exts):
yield (i, os.path.relpath(fpath, root), 0)
i += 1
+
def write_list(path_out, image_list):
with open(path_out, 'w') as fout:
for i, item in enumerate(image_list):
@@ -69,6 +75,7 @@ def write_list(path_out, image_list):
line += '%s\n' % item[1]
fout.write(line)
+
def make_list(args):
image_list = list_image(args.root, args.recursive, args.exts)
image_list = list(image_list)
@@ -94,6 +101,7 @@ def make_list(args):
write_list(args.prefix + str_chunk + '_val.lst', chunk[sep_test + sep:])
write_list(args.prefix + str_chunk + '_train.lst', chunk[sep_test:sep_test + sep])
+
def read_list(path_in):
with open(path_in) as fin:
while True:
@@ -103,15 +111,16 @@ def read_list(path_in):
line = [i.strip() for i in line.strip().split('\t')]
line_len = len(line)
if line_len < 3:
- print('lst should at least has three parts, but only has %s parts for %s' %(line_len, line))
+ print('lst should at least has three parts, but only has %s parts for %s' % (line_len, line))
continue
try:
item = [int(line[0])] + [line[-1]] + [float(i) for i in line[1:-1]]
except Exception as e:
- print('Parsing lst met error for %s, detail: %s' %(line, e))
+ print('Parsing lst met error for %s, detail: %s' % (line, e))
continue
yield item
+
def image_encode(args, i, item, q_out):
fullpath = os.path.join(args.root, item[1])
@@ -145,10 +154,10 @@ def image_encode(args, i, item, q_out):
return
if args.center_crop:
if img.shape[0] > img.shape[1]:
- margin = (img.shape[0] - img.shape[1]) // 2;
+ margin = (img.shape[0] - img.shape[1]) // 2
img = img[margin:margin + img.shape[1], :]
else:
- margin = (img.shape[1] - img.shape[0]) // 2;
+ margin = (img.shape[1] - img.shape[0]) // 2
img = img[:, margin:margin + img.shape[0]]
if args.resize:
if img.shape[0] > img.shape[1]:
@@ -166,6 +175,7 @@ def image_encode(args, i, item, q_out):
q_out.put((i, None, item))
return
+
def read_worker(args, q_in, q_out):
while True:
deq = q_in.get()
@@ -174,6 +184,7 @@ def read_worker(args, q_in, q_out):
i, item = deq
image_encode(args, i, item, q_out)
+
def write_worker(q_out, fname, working_dir):
pre_time = time.time()
count = 0
@@ -203,6 +214,7 @@ def write_worker(q_out, fname, working_dir):
pre_time = cur_time
count += 1
+
def parse_args():
parser = argparse.ArgumentParser(
formatter_class=argparse.ArgumentDefaultsHelpFormatter,
@@ -252,12 +264,13 @@ def parse_args():
rgroup.add_argument('--encoding', type=str, default='.jpg', choices=['.jpg', '.png'],
help='specify the encoding of the images.')
rgroup.add_argument('--pack-label', type=bool, default=False,
- help='Whether to also pack multi dimensional label in the record file')
+ help='Whether to also pack multi dimensional label in the record file')
args = parser.parse_args()
args.prefix = os.path.abspath(args.prefix)
args.root = os.path.abspath(args.root)
return args
+
if __name__ == '__main__':
args = parse_args()
if args.list:
@@ -268,7 +281,7 @@ def parse_args():
else:
working_dir = os.path.dirname(args.prefix)
files = [os.path.join(working_dir, fname) for fname in os.listdir(working_dir)
- if os.path.isfile(os.path.join(working_dir, fname))]
+ if os.path.isfile(os.path.join(working_dir, fname))]
count = 0
for fname in files:
if fname.startswith(args.prefix) and fname.endswith('.lst'):
@@ -279,7 +292,7 @@ def parse_args():
if args.num_thread > 1 and multiprocessing is not None:
q_in = [multiprocessing.Queue(1024) for i in range(args.num_thread)]
q_out = multiprocessing.Queue(1024)
- read_process = [multiprocessing.Process(target=read_worker, args=(args, q_in[i], q_out)) \
+ read_process = [multiprocessing.Process(target=read_worker, args=(args, q_in[i], q_out))
for i in range(args.num_thread)]
for p in read_process:
p.start()
@@ -321,4 +334,4 @@ def parse_args():
pre_time = cur_time
cnt += 1
if not count:
- print('Did not find and list file with prefix %s'%args.prefix)
+ print('Did not find and list file with prefix %s' % args.prefix)
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
With regards,
Apache Git Services