yujunjun/PaddleOCR
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/PaddleOCR.git synced 2025-11-23 05:26:19 +00:00
Code Issues Packages Projects Releases Wiki Activity

rename rec_resnet_fpn

Browse Source
This commit is contained in:
tink2123 2020-08-26 20:14:13 +08:00
parent 97cfef3265
commit 5edb619cdd
2 changed files with 98 additions and 41 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
configs/rec/rec_r50fpn_vd_none_srn.yml
View File

@ -27,7 +27,7 @@ Architecture:
function: ppocr.modeling.architectures.rec_model,RecModel function: ppocr.modeling.architectures.rec_model,RecModel
Backbone: Backbone:
function: ppocr.modeling.backbones.rec_resnet50_fpn,ResNet function: ppocr.modeling.backbones.rec_resnet_fpn,ResNet
layers: 50 layers: 50
Head: Head:

117
ppocr/modeling/backbones/rec_resnet50_fpn.py → ppocr/modeling/backbones/rec_resnet_fpn.py
View File

@ -22,12 +22,12 @@ import paddle
import paddle.fluid as fluid import paddle.fluid as fluid
from paddle.fluid.param_attr import ParamAttr from paddle.fluid.param_attr import ParamAttr
__all__ = [
__all__ = ["ResNet", "ResNet18", "ResNet34", "ResNet50", "ResNet101", "ResNet152"] "ResNet", "ResNet18", "ResNet34", "ResNet50", "ResNet101", "ResNet152"
]
Trainable = True Trainable = True
w_nolr = fluid.ParamAttr( w_nolr = fluid.ParamAttr(trainable=Trainable)
trainable = Trainable)
train_parameters = { train_parameters = {
"input_size": [3, 224, 224], "input_size": [3, 224, 224],
"input_mean": [0.485, 0.456, 0.406], "input_mean": [0.485, 0.456, 0.406],
@ -40,12 +40,12 @@ train_parameters = {
} }
} }
class ResNet(): class ResNet():
def __init__(self, params): def __init__(self, params):
self.layers = params['layers'] self.layers = params['layers']
self.params = train_parameters self.params = train_parameters
def __call__(self, input): def __call__(self, input):
layers = self.layers layers = self.layers
supported_layers = [18, 34, 50, 101, 152] supported_layers = [18, 34, 50, 101, 152]
@ -60,11 +60,16 @@ class ResNet():
depth = [3, 4, 23, 3] depth = [3, 4, 23, 3]
elif layers == 152: elif layers == 152:
depth = [3, 8, 36, 3] depth = [3, 8, 36, 3]
stride_list = [(2,2),(2,2),(1,1),(1,1)] stride_list = [(2, 2), (2, 2), (1, 1), (1, 1)]
num_filters = [64, 128, 256, 512] num_filters = [64, 128, 256, 512]
conv = self.conv_bn_layer( conv = self.conv_bn_layer(
input=input, num_filters=64, filter_size=7, stride=2, act='relu', name="conv1") input=input,
num_filters=64,
filter_size=7,
stride=2,
act='relu',
name="conv1")
F = [] F = []
if layers >= 50: if layers >= 50:
for block in range(len(depth)): for block in range(len(depth)):
@ -79,7 +84,8 @@ class ResNet():
conv = self.bottleneck_block( conv = self.bottleneck_block(
input=conv, input=conv,
num_filters=num_filters[block], num_filters=num_filters[block],
stride=stride_list[block] if i == 0 else 1, name=conv_name) stride=stride_list[block] if i == 0 else 1,
name=conv_name)
F.append(conv) F.append(conv)
else: else:
for block in range(len(depth)): for block in range(len(depth)):
@ -102,20 +108,43 @@ class ResNet():
base = F[-1] base = F[-1]
for i in [-2, -3]: for i in [-2, -3]:
b, c, w, h = F[i].shape b, c, w, h = F[i].shape
if (w,h) == base.shape[2:]: if (w, h) == base.shape[2:]:
base = base base = base
else: else:
base = fluid.layers.conv2d_transpose( input=base, num_filters=c,filter_size=4, stride=2, base = fluid.layers.conv2d_transpose(
padding=1,act=None, input=base,
num_filters=c,
filter_size=4,
stride=2,
padding=1,
act=None,
param_attr=w_nolr, param_attr=w_nolr,
bias_attr=w_nolr) bias_attr=w_nolr)
base = fluid.layers.batch_norm(base, act = "relu", param_attr=w_nolr, bias_attr=w_nolr) base = fluid.layers.batch_norm(
base, act="relu", param_attr=w_nolr, bias_attr=w_nolr)
base = fluid.layers.concat([base, F[i]], axis=1) base = fluid.layers.concat([base, F[i]], axis=1)
base = fluid.layers.conv2d(base, num_filters=c, filter_size=1, param_attr=w_nolr, bias_attr=w_nolr) base = fluid.layers.conv2d(
base = fluid.layers.conv2d(base, num_filters=c, filter_size=3,padding = 1, param_attr=w_nolr, bias_attr=w_nolr) base,
base = fluid.layers.batch_norm(base, act = "relu", param_attr=w_nolr, bias_attr=w_nolr) num_filters=c,
filter_size=1,
param_attr=w_nolr,
bias_attr=w_nolr)
base = fluid.layers.conv2d(
base,
num_filters=c,
filter_size=3,
padding=1,
param_attr=w_nolr,
bias_attr=w_nolr)
base = fluid.layers.batch_norm(
base, act="relu", param_attr=w_nolr, bias_attr=w_nolr)
base = fluid.layers.conv2d(base, num_filters=512, filter_size=1,bias_attr=w_nolr,param_attr=w_nolr) base = fluid.layers.conv2d(
base,
num_filters=512,
filter_size=1,
bias_attr=w_nolr,
param_attr=w_nolr)
return base return base
@ -130,13 +159,14 @@ class ResNet():
conv = fluid.layers.conv2d( conv = fluid.layers.conv2d(
input=input, input=input,
num_filters=num_filters, num_filters=num_filters,
filter_size= 2 if stride==(1,1) else filter_size, filter_size=2 if stride == (1, 1) else filter_size,
dilation = 2 if stride==(1,1) else 1, dilation=2 if stride == (1, 1) else 1,
stride=stride, stride=stride,
padding=(filter_size - 1) // 2, padding=(filter_size - 1) // 2,
groups=groups, groups=groups,
act=None, act=None,
param_attr=ParamAttr(name=name + "_weights",trainable = Trainable), param_attr=ParamAttr(
name=name + "_weights", trainable=Trainable),
bias_attr=False, bias_attr=False,
name=name + '.conv2d.output.1') name=name + '.conv2d.output.1')
@ -144,18 +174,21 @@ class ResNet():
bn_name = "bn_" + name bn_name = "bn_" + name
else: else:
bn_name = "bn" + name[3:] bn_name = "bn" + name[3:]
return fluid.layers.batch_norm(input=conv, return fluid.layers.batch_norm(
input=conv,
act=act, act=act,
name=bn_name + '.output.1', name=bn_name + '.output.1',
param_attr=ParamAttr(name=bn_name + '_scale',trainable = Trainable), param_attr=ParamAttr(
bias_attr=ParamAttr(bn_name + '_offset',trainable = Trainable), name=bn_name + '_scale', trainable=Trainable),
bias_attr=ParamAttr(
bn_name + '_offset', trainable=Trainable),
moving_mean_name=bn_name + '_mean', moving_mean_name=bn_name + '_mean',
moving_variance_name=bn_name + '_variance', ) moving_variance_name=bn_name + '_variance', )
def shortcut(self, input, ch_out, stride, is_first, name): def shortcut(self, input, ch_out, stride, is_first, name):
ch_in = input.shape[1] ch_in = input.shape[1]
if ch_in != ch_out or stride != 1 or is_first == True: if ch_in != ch_out or stride != 1 or is_first == True:
if stride == (1,1): if stride == (1, 1):
return self.conv_bn_layer(input, ch_out, 1, 1, name=name) return self.conv_bn_layer(input, ch_out, 1, 1, name=name)
else: #stride == (2,2) else: #stride == (2,2)
return self.conv_bn_layer(input, ch_out, 1, stride, name=name) return self.conv_bn_layer(input, ch_out, 1, stride, name=name)
@ -165,7 +198,11 @@ class ResNet():
def bottleneck_block(self, input, num_filters, stride, name): def bottleneck_block(self, input, num_filters, stride, name):
conv0 = self.conv_bn_layer( conv0 = self.conv_bn_layer(
input=input, num_filters=num_filters, filter_size=1, act='relu', name=name + "_branch2a") input=input,
num_filters=num_filters,
filter_size=1,
act='relu',
name=name + "_branch2a")
conv1 = self.conv_bn_layer( conv1 = self.conv_bn_layer(
input=conv0, input=conv0,
num_filters=num_filters, num_filters=num_filters,
@ -174,16 +211,36 @@ class ResNet():
act='relu', act='relu',
name=name + "_branch2b") name=name + "_branch2b")
conv2 = self.conv_bn_layer( conv2 = self.conv_bn_layer(
input=conv1, num_filters=num_filters * 4, filter_size=1, act=None, name=name + "_branch2c") input=conv1,
num_filters=num_filters * 4,
filter_size=1,
act=None,
name=name + "_branch2c")
short = self.shortcut(input, num_filters * 4, stride, is_first=False, name=name + "_branch1") short = self.shortcut(
input,
num_filters * 4,
stride,
is_first=False,
name=name + "_branch1")
return fluid.layers.elementwise_add(x=short, y=conv2, act='relu', name=name + ".add.output.5") return fluid.layers.elementwise_add(
x=short, y=conv2, act='relu', name=name + ".add.output.5")
def basic_block(self, input, num_filters, stride, is_first, name): def basic_block(self, input, num_filters, stride, is_first, name):
conv0 = self.conv_bn_layer(input=input, num_filters=num_filters, filter_size=3, act='relu', stride=stride, conv0 = self.conv_bn_layer(
input=input,
num_filters=num_filters,
filter_size=3,
act='relu',
stride=stride,
name=name + "_branch2a") name=name + "_branch2a")
conv1 = self.conv_bn_layer(input=conv0, num_filters=num_filters, filter_size=3, act=None, conv1 = self.conv_bn_layer(
input=conv0,
num_filters=num_filters,
filter_size=3,
act=None,
name=name + "_branch2b") name=name + "_branch2b")
short = self.shortcut(input, num_filters, stride, is_first, name=name + "_branch1") short = self.shortcut(
input, num_filters, stride, is_first, name=name + "_branch1")
return fluid.layers.elementwise_add(x=short, y=conv1, act='relu') return fluid.layers.elementwise_add(x=short, y=conv1, act='relu')