Skip to content

P
PaddleSeg

PaddlePaddle / PaddleSeg

通知 285
Star 8
Fork 1
P
PaddleSeg
提交 8f75ee50 编写于 作者: M michaelowenliu
浏览文件
操作

Merge branch 'develop' of https://github.com/PaddlePaddle/PaddleSeg into develop

上级 858077ef 71c4b688
隐藏空白更改
内联 并排
Showing with 242 addition and 346 deletion
dygraph/configs/_base_/cityscapes.yml
浏览文件 @ 8f75ee50
...@@ -32,6 +32,7 @@ learning_rate: ...@@ -32,6 +32,7 @@ learning_rate:
decay: decay:
type: poly type: poly
power: 0.9 power: 0.9
end_lr: 0.0
loss: loss:
types: types:
......
dygraph/configs/_base_/optic_disc_seg.yml
浏览文件 @ 8f75ee50
...@@ -30,6 +30,7 @@ learning_rate: ...@@ -30,6 +30,7 @@ learning_rate:
decay: decay:
type: poly type: poly
power: 0.9 power: 0.9
end_lr: 0
loss: loss:
types: types:
......
dygraph/configs/fcn_hrnet/fcn_hrnetw18_cityscapes_1024x512_100k.yml
浏览文件 @ 8f75ee50
...@@ -4,9 +4,10 @@ model: ...@@ -4,9 +4,10 @@ model:
type: FCN type: FCN
backbone: backbone:
type: HRNet_W18 type: HRNet_W18
pretrained: pretrained_model/hrnet_w18_imagenet
num_classes: 19 num_classes: 19
backbone_channels: [270] pretrained: Null
backbone_pretrained: pretrained_model/hrnet_w18_imagenet backbone_indices: [-1]
optimizer: optimizer:
weight_decay: 0.0005 weight_decay: 0.0005
dygraph/configs/fcn_hrnet/fcn_hrnetw18_optic_disc_512x512_10k.yml
浏览文件 @ 8f75ee50
...@@ -4,6 +4,7 @@ model: ...@@ -4,6 +4,7 @@ model:
type: FCN type: FCN
backbone: backbone:
type: HRNet_W18 type: HRNet_W18
num_classes: 2 pretrained: pretrained_model/hrnet_w18_imagenet
backbone_channels: [270] num_classes: 19
backbone_pretrained: pretrained_model/hrnet_w18_imagenet pretrained: Null
backbone_indices: [-1]
dygraph/configs/fcn_hrnet/fcn_hrnetw48_cityscapes_1024x512_100k.yml
浏览文件 @ 8f75ee50
...@@ -4,6 +4,10 @@ model: ...@@ -4,6 +4,10 @@ model:
type: FCN type: FCN
backbone: backbone:
type: HRNet_W48 type: HRNet_W48
pretrained: pretrained_model/hrnet_w48_imagenet
num_classes: 19 num_classes: 19
backbone_channels: [720] pretrained: Null
backbone_pretrained: pretrained_model/hrnet_w48_imagenet backbone_indices: [-1]
optimizer:
weight_decay: 0.0005
dygraph/configs/unet/unet_cityscapes_1024x512_40k.yml 0 → 100644
浏览文件 @ 8f75ee50
_base_: '../_base_/cityscapes.yml'
batch_size: 2
iters: 40000
model:
type: UNet
num_classes: 19
pretrained: Null
dygraph/paddleseg/models/backbones/hrnet.py
浏览文件 @ 8f75ee50
...@@ -16,16 +16,17 @@ import math ...@@ -16,16 +16,17 @@ import math
import os import os
import paddle import paddle
import paddle.fluid as fluid from paddle import ParamAttr
from paddle.fluid.param_attr import ParamAttr import paddle.nn as nn
from paddle.fluid.layer_helper import LayerHelper import paddle.nn.functional as F
from paddle.fluid.dygraph.nn import Conv2D, Pool2D, Linear
from paddle.fluid.initializer import Normal
from paddle.nn import SyncBatchNorm as BatchNorm from paddle.nn import SyncBatchNorm as BatchNorm
from paddle.nn import Conv2d, Linear
from paddle.nn import AdaptiveAvgPool2d, MaxPool2d, AvgPool2d
from paddleseg.cvlibs import manager from paddleseg.cvlibs import manager
from paddleseg.utils import utils from paddleseg.utils import utils
from paddleseg.cvlibs import param_init from paddleseg.cvlibs import param_init
from paddleseg.models.common import layer_libs
__all__ = [ __all__ = [
"HRNet_W18_Small_V1", "HRNet_W18_Small_V2", "HRNet_W18", "HRNet_W30", "HRNet_W18_Small_V1", "HRNet_W18_Small_V2", "HRNet_W18", "HRNet_W30",
...@@ -33,7 +34,7 @@ __all__ = [ ...@@ -33,7 +34,7 @@ __all__ = [
] ]
class HRNet(fluid.dygraph.Layer): class HRNet(nn.Layer):
""" """
HRNet：Deep High-Resolution Representation Learning for Visual Recognition HRNet：Deep High-Resolution Representation Learning for Visual Recognition
https://arxiv.org/pdf/1908.07919.pdf. https://arxiv.org/pdf/1908.07919.pdf.
...@@ -56,6 +57,7 @@ class HRNet(fluid.dygraph.Layer): ...@@ -56,6 +57,7 @@ class HRNet(fluid.dygraph.Layer):
""" """
def __init__(self, def __init__(self,
pretrained=None,
stage1_num_modules=1, stage1_num_modules=1,
stage1_num_blocks=[4], stage1_num_blocks=[4],
stage1_num_channels=[64], stage1_num_channels=[64],
...@@ -70,7 +72,7 @@ class HRNet(fluid.dygraph.Layer): ...@@ -70,7 +72,7 @@ class HRNet(fluid.dygraph.Layer):
stage4_num_channels=[18, 36, 72, 144], stage4_num_channels=[18, 36, 72, 144],
has_se=False): has_se=False):
super(HRNet, self).__init__() super(HRNet, self).__init__()
self.pretrained = pretrained
self.stage1_num_modules = stage1_num_modules self.stage1_num_modules = stage1_num_modules
self.stage1_num_blocks = stage1_num_blocks self.stage1_num_blocks = stage1_num_blocks
self.stage1_num_channels = stage1_num_channels self.stage1_num_channels = stage1_num_channels
...@@ -84,22 +86,23 @@ class HRNet(fluid.dygraph.Layer): ...@@ -84,22 +86,23 @@ class HRNet(fluid.dygraph.Layer):
self.stage4_num_blocks = stage4_num_blocks self.stage4_num_blocks = stage4_num_blocks
self.stage4_num_channels = stage4_num_channels self.stage4_num_channels = stage4_num_channels
self.has_se = has_se self.has_se = has_se
self.feat_channels = [sum(stage4_num_channels)]
self.conv_layer1_1 = ConvBNLayer( self.conv_layer1_1 = layer_libs.ConvBNReLU(
num_channels=3, in_channels=3,
num_filters=64, out_channels=64,
filter_size=3, kernel_size=3,
stride=2, stride=2,
act='relu', padding='same',
name="layer1_1") bias_attr=False)
self.conv_layer1_2 = ConvBNLayer( self.conv_layer1_2 = layer_libs.ConvBNReLU(
num_channels=64, in_channels=64,
num_filters=64, out_channels=64,
filter_size=3, kernel_size=3,
stride=2, stride=2,
act='relu', padding='same',
name="layer1_2") bias_attr=False)
self.la1 = Layer1( self.la1 = Layer1(
num_channels=64, num_channels=64,
...@@ -144,6 +147,7 @@ class HRNet(fluid.dygraph.Layer): ...@@ -144,6 +147,7 @@ class HRNet(fluid.dygraph.Layer):
num_filters=self.stage4_num_channels, num_filters=self.stage4_num_channels,
has_se=self.has_se, has_se=self.has_se,
name="st4") name="st4")
self.init_weight()
def forward(self, x, label=None, mode='train'): def forward(self, x, label=None, mode='train'):
input_shape = x.shape[2:] input_shape = x.shape[2:]
...@@ -162,45 +166,29 @@ class HRNet(fluid.dygraph.Layer): ...@@ -162,45 +166,29 @@ class HRNet(fluid.dygraph.Layer):
st4 = self.st4(tr3) st4 = self.st4(tr3)
x0_h, x0_w = st4[0].shape[2:] x0_h, x0_w = st4[0].shape[2:]
x1 = fluid.layers.resize_bilinear(st4[1], out_shape=(x0_h, x0_w)) x1 = F.resize_bilinear(st4[1], out_shape=(x0_h, x0_w))
x2 = fluid.layers.resize_bilinear(st4[2], out_shape=(x0_h, x0_w)) x2 = F.resize_bilinear(st4[2], out_shape=(x0_h, x0_w))
x3 = fluid.layers.resize_bilinear(st4[3], out_shape=(x0_h, x0_w)) x3 = F.resize_bilinear(st4[3], out_shape=(x0_h, x0_w))
x = fluid.layers.concat([st4[0], x1, x2, x3], axis=1) x = paddle.concat([st4[0], x1, x2, x3], axis=1)
return [x] return [x]
def init_weight(self):
class ConvBNLayer(fluid.dygraph.Layer): params = self.parameters()
def __init__(self, for param in params:
num_channels, param_name = param.name
num_filters, if 'batch_norm' in param_name:
filter_size, if 'w_0' in param_name:
stride=1, param_init.constant_init(param, value=1.0)
groups=1, elif 'b_0' in param_name:
act="relu", param_init.constant_init(param, value=0.0)
name=None): if 'conv' in param_name and 'w_0' in param_name:
super(ConvBNLayer, self).__init__() param_init.normal_init(param, scale=0.001)
if self.pretrained is not None:
self._conv = Conv2D( utils.load_pretrained_model(self, self.pretrained)
num_channels=num_channels,
num_filters=num_filters,
filter_size=filter_size, class Layer1(nn.Layer):
stride=stride,
padding=(filter_size - 1) // 2,
groups=groups,
bias_attr=False)
self._batch_norm = BatchNorm(num_filters)
self.act = act
def forward(self, input):
y = self._conv(input)
y = self._batch_norm(y)
if self.act == 'relu':
y = fluid.layers.relu(y)
return y
class Layer1(fluid.dygraph.Layer):
def __init__(self, def __init__(self,
num_channels, num_channels,
num_filters, num_filters,
...@@ -230,7 +218,7 @@ class Layer1(fluid.dygraph.Layer): ...@@ -230,7 +218,7 @@ class Layer1(fluid.dygraph.Layer):
return conv return conv
class TransitionLayer(fluid.dygraph.Layer): class TransitionLayer(nn.Layer):
def __init__(self, in_channels, out_channels, name=None): def __init__(self, in_channels, out_channels, name=None):
super(TransitionLayer, self).__init__() super(TransitionLayer, self).__init__()
...@@ -243,20 +231,22 @@ class TransitionLayer(fluid.dygraph.Layer): ...@@ -243,20 +231,22 @@ class TransitionLayer(fluid.dygraph.Layer):
if in_channels[i] != out_channels[i]: if in_channels[i] != out_channels[i]:
residual = self.add_sublayer( residual = self.add_sublayer(
"transition_{}_layer_{}".format(name, i + 1), "transition_{}_layer_{}".format(name, i + 1),
ConvBNLayer( layer_libs.ConvBNReLU(
num_channels=in_channels[i], in_channels=in_channels[i],
num_filters=out_channels[i], out_channels=out_channels[i],
filter_size=3, kernel_size=3,
name=name + '_layer_' + str(i + 1))) padding='same',
bias_attr=False))
else: else:
residual = self.add_sublayer( residual = self.add_sublayer(
"transition_{}_layer_{}".format(name, i + 1), "transition_{}_layer_{}".format(name, i + 1),
ConvBNLayer( layer_libs.ConvBNReLU(
num_channels=in_channels[-1], in_channels=in_channels[-1],
num_filters=out_channels[i], out_channels=out_channels[i],
filter_size=3, kernel_size=3,
stride=2, stride=2,
name=name + '_layer_' + str(i + 1))) padding='same',
bias_attr=False))
self.conv_bn_func_list.append(residual) self.conv_bn_func_list.append(residual)
def forward(self, input): def forward(self, input):
...@@ -272,7 +262,7 @@ class TransitionLayer(fluid.dygraph.Layer): ...@@ -272,7 +262,7 @@ class TransitionLayer(fluid.dygraph.Layer):
return outs return outs
class Branches(fluid.dygraph.Layer): class Branches(nn.Layer):
def __init__(self, def __init__(self,
num_blocks, num_blocks,
in_channels, in_channels,
...@@ -307,7 +297,7 @@ class Branches(fluid.dygraph.Layer): ...@@ -307,7 +297,7 @@ class Branches(fluid.dygraph.Layer):
return outs return outs
class BottleneckBlock(fluid.dygraph.Layer): class BottleneckBlock(nn.Layer):
def __init__(self, def __init__(self,
num_channels, num_channels,
num_filters, num_filters,
...@@ -320,34 +310,35 @@ class BottleneckBlock(fluid.dygraph.Layer): ...@@ -320,34 +310,35 @@ class BottleneckBlock(fluid.dygraph.Layer):
self.has_se = has_se self.has_se = has_se
self.downsample = downsample self.downsample = downsample
self.conv1 = ConvBNLayer( self.conv1 = layer_libs.ConvBNReLU(
num_channels=num_channels, in_channels=num_channels,
num_filters=num_filters, out_channels=num_filters,
filter_size=1, kernel_size=1,
act="relu", padding='same',
name=name + "_conv1", bias_attr=False)
)
self.conv2 = ConvBNLayer( self.conv2 = layer_libs.ConvBNReLU(
num_channels=num_filters, in_channels=num_filters,
num_filters=num_filters, out_channels=num_filters,
filter_size=3, kernel_size=3,
stride=stride, stride=stride,
act="relu", padding='same',
name=name + "_conv2") bias_attr=False)
self.conv3 = ConvBNLayer(
num_channels=num_filters, self.conv3 = layer_libs.ConvBN(
num_filters=num_filters * 4, in_channels=num_filters,
filter_size=1, out_channels=num_filters * 4,
act=None, kernel_size=1,
name=name + "_conv3") padding='same',
bias_attr=False)
if self.downsample: if self.downsample:
self.conv_down = ConvBNLayer( self.conv_down = layer_libs.ConvBN(
num_channels=num_channels, in_channels=num_channels,
num_filters=num_filters * 4, out_channels=num_filters * 4,
filter_size=1, kernel_size=1,
act=None, padding='same',
name=name + "_downsample") bias_attr=False)
if self.has_se: if self.has_se:
self.se = SELayer( self.se = SELayer(
...@@ -368,11 +359,12 @@ class BottleneckBlock(fluid.dygraph.Layer): ...@@ -368,11 +359,12 @@ class BottleneckBlock(fluid.dygraph.Layer):
if self.has_se: if self.has_se:
conv3 = self.se(conv3) conv3 = self.se(conv3)
y = fluid.layers.elementwise_add(x=conv3, y=residual, act="relu") y = conv3 + residual
y = F.relu(y)
return y return y
class BasicBlock(fluid.dygraph.Layer): class BasicBlock(nn.Layer):
def __init__(self, def __init__(self,
num_channels, num_channels,
num_filters, num_filters,
...@@ -385,28 +377,27 @@ class BasicBlock(fluid.dygraph.Layer): ...@@ -385,28 +377,27 @@ class BasicBlock(fluid.dygraph.Layer):
self.has_se = has_se self.has_se = has_se
self.downsample = downsample self.downsample = downsample
self.conv1 = ConvBNLayer( self.conv1 = layer_libs.ConvBNReLU(
num_channels=num_channels, in_channels=num_channels,
num_filters=num_filters, out_channels=num_filters,
filter_size=3, kernel_size=3,
stride=stride, stride=stride,
act="relu", padding='same',
name=name + "_conv1") bias_attr=False)
self.conv2 = ConvBNLayer( self.conv2 = layer_libs.ConvBN(
num_channels=num_filters, in_channels=num_filters,
num_filters=num_filters, out_channels=num_filters,
filter_size=3, kernel_size=3,
stride=1, padding='same',
act=None, bias_attr=False)
name=name + "_conv2")
if self.downsample: if self.downsample:
self.conv_down = ConvBNLayer( self.conv_down = layer_libs.ConvBNReLU(
num_channels=num_channels, in_channels=num_channels,
num_filters=num_filters * 4, out_channels=num_filters,
filter_size=1, kernel_size=1,
act="relu", padding='same',
name=name + "_downsample") bias_attr=False)
if self.has_se: if self.has_se:
self.se = SELayer( self.se = SELayer(
...@@ -426,15 +417,16 @@ class BasicBlock(fluid.dygraph.Layer): ...@@ -426,15 +417,16 @@ class BasicBlock(fluid.dygraph.Layer):
if self.has_se: if self.has_se:
conv2 = self.se(conv2) conv2 = self.se(conv2)
y = fluid.layers.elementwise_add(x=conv2, y=residual, act="relu") y = conv2 + residual
y = F.relu(y)
return y return y
class SELayer(fluid.dygraph.Layer): class SELayer(nn.Layer):
def __init__(self, num_channels, num_filters, reduction_ratio, name=None): def __init__(self, num_channels, num_filters, reduction_ratio, name=None):
super(SELayer, self).__init__() super(SELayer, self).__init__()
self.pool2d_gap = Pool2D(pool_type='avg', global_pooling=True) self.pool2d_gap = AdaptiveAvgPool2d(1)
self._num_channels = num_channels self._num_channels = num_channels
...@@ -445,9 +437,7 @@ class SELayer(fluid.dygraph.Layer): ...@@ -445,9 +437,7 @@ class SELayer(fluid.dygraph.Layer):
med_ch, med_ch,
act="relu", act="relu",
param_attr=ParamAttr( param_attr=ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv), initializer=nn.initializer.Uniform(-stdv, stdv)))
name=name + "_sqz_weights"),
bias_attr=ParamAttr(name=name + '_sqz_offset'))
stdv = 1.0 / math.sqrt(med_ch * 1.0) stdv = 1.0 / math.sqrt(med_ch * 1.0)
self.excitation = Linear( self.excitation = Linear(
...@@ -455,22 +445,20 @@ class SELayer(fluid.dygraph.Layer): ...@@ -455,22 +445,20 @@ class SELayer(fluid.dygraph.Layer):
num_filters, num_filters,
act="sigmoid", act="sigmoid",
param_attr=ParamAttr( param_attr=ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv), initializer=nn.initializer.Uniform(-stdv, stdv)))
name=name + "_exc_weights"),
bias_attr=ParamAttr(name=name + '_exc_offset'))
def forward(self, input): def forward(self, input):
pool = self.pool2d_gap(input) pool = self.pool2d_gap(input)
pool = fluid.layers.reshape(pool, shape=[-1, self._num_channels]) pool = paddle.reshape(pool, shape=[-1, self._num_channels])
squeeze = self.squeeze(pool) squeeze = self.squeeze(pool)
excitation = self.excitation(squeeze) excitation = self.excitation(squeeze)
excitation = fluid.layers.reshape( excitation = paddle.reshape(
excitation, shape=[-1, self._num_channels, 1, 1]) excitation, shape=[-1, self._num_channels, 1, 1])
out = input * excitation out = input * excitation
return out return out
class Stage(fluid.dygraph.Layer): class Stage(nn.Layer):
def __init__(self, def __init__(self,
num_channels, num_channels,
num_modules, num_modules,
...@@ -514,7 +502,7 @@ class Stage(fluid.dygraph.Layer): ...@@ -514,7 +502,7 @@ class Stage(fluid.dygraph.Layer):
return out return out
class HighResolutionModule(fluid.dygraph.Layer): class HighResolutionModule(nn.Layer):
def __init__(self, def __init__(self,
num_channels, num_channels,
num_blocks, num_blocks,
...@@ -543,7 +531,7 @@ class HighResolutionModule(fluid.dygraph.Layer): ...@@ -543,7 +531,7 @@ class HighResolutionModule(fluid.dygraph.Layer):
return out return out
class FuseLayers(fluid.dygraph.Layer): class FuseLayers(nn.Layer):
def __init__(self, def __init__(self,
in_channels, in_channels,
out_channels, out_channels,
...@@ -561,14 +549,12 @@ class FuseLayers(fluid.dygraph.Layer): ...@@ -561,14 +549,12 @@ class FuseLayers(fluid.dygraph.Layer):
if j > i: if j > i:
residual_func = self.add_sublayer( residual_func = self.add_sublayer(
"residual_{}_layer_{}_{}".format(name, i + 1, j + 1), "residual_{}_layer_{}_{}".format(name, i + 1, j + 1),
ConvBNLayer( layer_libs.ConvBN(
num_channels=in_channels[j], in_channels=in_channels[j],
num_filters=out_channels[i], out_channels=out_channels[i],
filter_size=1, kernel_size=1,
stride=1, padding='same',
act=None, bias_attr=False))
name=name + '_layer_' + str(i + 1) + '_' +
str(j + 1)))
self.residual_func_list.append(residual_func) self.residual_func_list.append(residual_func)
elif j < i: elif j < i:
pre_num_filters = in_channels[j] pre_num_filters = in_channels[j]
...@@ -577,27 +563,25 @@ class FuseLayers(fluid.dygraph.Layer): ...@@ -577,27 +563,25 @@ class FuseLayers(fluid.dygraph.Layer):
residual_func = self.add_sublayer( residual_func = self.add_sublayer(
"residual_{}_layer_{}_{}_{}".format( "residual_{}_layer_{}_{}_{}".format(
name, i + 1, j + 1, k + 1), name, i + 1, j + 1, k + 1),
ConvBNLayer( layer_libs.ConvBN(
num_channels=pre_num_filters, in_channels=pre_num_filters,
num_filters=out_channels[i], out_channels=out_channels[i],
filter_size=3, kernel_size=3,
stride=2, stride=2,
act=None, padding='same',
name=name + '_layer_' + str(i + 1) + '_' + bias_attr=False))
str(j + 1) + '_' + str(k + 1)))
pre_num_filters = out_channels[i] pre_num_filters = out_channels[i]
else: else:
residual_func = self.add_sublayer( residual_func = self.add_sublayer(
"residual_{}_layer_{}_{}_{}".format( "residual_{}_layer_{}_{}_{}".format(
name, i + 1, j + 1, k + 1), name, i + 1, j + 1, k + 1),
ConvBNLayer( layer_libs.ConvBNReLU(
num_channels=pre_num_filters, in_channels=pre_num_filters,
num_filters=out_channels[j], out_channels=out_channels[j],
filter_size=3, kernel_size=3,
stride=2, stride=2,
act="relu", padding='same',
name=name + '_layer_' + str(i + 1) + '_' + bias_attr=False))
str(j + 1) + '_' + str(k + 1)))
pre_num_filters = out_channels[j] pre_num_filters = out_channels[j]
self.residual_func_list.append(residual_func) self.residual_func_list.append(residual_func)
...@@ -612,54 +596,22 @@ class FuseLayers(fluid.dygraph.Layer): ...@@ -612,54 +596,22 @@ class FuseLayers(fluid.dygraph.Layer):
y = self.residual_func_list[residual_func_idx](input[j]) y = self.residual_func_list[residual_func_idx](input[j])
residual_func_idx += 1 residual_func_idx += 1
y = fluid.layers.resize_bilinear( y = F.resize_bilinear(input=y, out_shape=residual_shape)
input=y, out_shape=residual_shape) residual = residual + y
residual = fluid.layers.elementwise_add(
x=residual, y=y, act=None)
elif j < i: elif j < i:
y = input[j] y = input[j]
for k in range(i - j): for k in range(i - j):
y = self.residual_func_list[residual_func_idx](y) y = self.residual_func_list[residual_func_idx](y)
residual_func_idx += 1 residual_func_idx += 1
residual = fluid.layers.elementwise_add( residual = residual + y
x=residual, y=y, act=None)
layer_helper = LayerHelper(self.full_name(), act='relu') residual = F.relu(residual)
residual = layer_helper.append_activation(residual)
outs.append(residual) outs.append(residual)
return outs return outs
class LastClsOut(fluid.dygraph.Layer):
def __init__(self,
num_channel_list,
has_se,
num_filters_list=[32, 64, 128, 256],
name=None):
super(LastClsOut, self).__init__()
self.func_list = []
for idx in range(len(num_channel_list)):
func = self.add_sublayer(
"conv_{}_conv_{}".format(name, idx + 1),
BottleneckBlock(
num_channels=num_channel_list[idx],
num_filters=num_filters_list[idx],
has_se=has_se,
downsample=True,
name=name + 'conv_' + str(idx + 1)))
self.func_list.append(func)
def forward(self, inputs):
outs = []
for idx, input in enumerate(inputs):
out = self.func_list[idx](input)
outs.append(out)
return outs
@manager.BACKBONES.add_component @manager.BACKBONES.add_component
def HRNet_W18_Small_V1(**kwargs): def HRNet_W18_Small_V1(**kwargs):
model = HRNet( model = HRNet(
......
dygraph/paddleseg/models/fcn.py
浏览文件 @ 8f75ee50
...@@ -36,64 +36,78 @@ __all__ = [ ...@@ -36,64 +36,78 @@ __all__ = [
@manager.MODELS.add_component @manager.MODELS.add_component
class FCN(nn.Layer): class FCN(nn.Layer):
def __init__(self,
num_classes,
backbone,
pretrained=None,
backbone_indices=(-1, ),
channels=None):
super(FCN, self).__init__()
self.backbone = backbone
backbone_channels = [
backbone.feat_channels[i] for i in backbone_indices
]
self.head = FCNHead(num_classes, backbone_indices, backbone_channels,
channels)
utils.load_entire_model(self, pretrained)
def forward(self, input):
feat_list = self.backbone(input)
logit_list = self.head(feat_list)
return [
F.resize_bilinear(logit, input.shape[2:]) for logit in logit_list
]
class FCNHead(nn.Layer):
""" """
Fully Convolutional Networks for Semantic Segmentation. A simple implementation for Fully Convolutional Networks for Semantic Segmentation.
https://arxiv.org/abs/1411.4038 https://arxiv.org/abs/1411.4038
Args: Args:
num_classes (int): the unique number of target classes. num_classes (int): the unique number of target classes.
backbone (paddle.nn.Layer): backbone networks. backbone (paddle.nn.Layer): backbone networks.
model_pretrained (str): the path of pretrained model. model_pretrained (str): the path of pretrained model.
backbone_indices (tuple): one values in the tuple indicte the indices of output of backbone.Default -1. backbone_indices (tuple): one values in the tuple indicte the indices of output of backbone.Default -1.
backbone_channels (tuple): the same length with "backbone_indices". It indicates the channels of corresponding index. backbone_channels (tuple): the same length with "backbone_indices". It indicates the channels of corresponding index.
channels (int): channels after conv layer before the last one. channels (int): channels after conv layer before the last one.
""" """
def __init__(self, def __init__(self,
num_classes, num_classes,
backbone,
backbone_pretrained=None,
model_pretrained=None,
backbone_indices=(-1, ), backbone_indices=(-1, ),
backbone_channels=(270, ), backbone_channels=(270, ),
channels=None): channels=None):
super(FCN, self).__init__() super(FCNHead, self).__init__()
self.num_classes = num_classes self.num_classes = num_classes
self.backbone_pretrained = backbone_pretrained
self.model_pretrained = model_pretrained
self.backbone_indices = backbone_indices self.backbone_indices = backbone_indices
if channels is None: if channels is None:
channels = backbone_channels[0] channels = backbone_channels[0]
self.backbone = backbone self.conv_1 = layer_libs.ConvBNReLU(
self.conv_last_2 = ConvBNLayer(
in_channels=backbone_channels[0], in_channels=backbone_channels[0],
out_channels=channels, out_channels=channels,
kernel_size=1, kernel_size=1,
padding='same',
stride=1) stride=1)
self.conv_last_1 = Conv2d( self.cls = Conv2d(
in_channels=channels, in_channels=channels,
out_channels=self.num_classes, out_channels=self.num_classes,
kernel_size=1, kernel_size=1,
stride=1, stride=1,
padding=0) padding=0)
if self.training: self.init_weight()
self.init_weight()
def forward(self, feat_list):
def forward(self, x): logit_list = []
input_shape = x.shape[2:] x = feat_list[self.backbone_indices[0]]
fea_list = self.backbone(x) x = self.conv_1(x)
x = fea_list[self.backbone_indices[0]] logit = self.cls(x)
x = self.conv_last_2(x) logit_list.append(logit)
logit = self.conv_last_1(x) return logit_list
logit = F.resize_bilinear(logit, input_shape)
return [logit]
def init_weight(self): def init_weight(self):
params = self.parameters() params = self.parameters()
...@@ -107,50 +121,6 @@ class FCN(nn.Layer): ...@@ -107,50 +121,6 @@ class FCN(nn.Layer):
if 'conv' in param_name and 'w_0' in param_name: if 'conv' in param_name and 'w_0' in param_name:
param_init.normal_init(param, scale=0.001) param_init.normal_init(param, scale=0.001)
if self.model_pretrained is not None:
if os.path.exists(self.model_pretrained):
utils.load_pretrained_model(self, self.model_pretrained)
else:
raise Exception('Pretrained model is not found: {}'.format(
self.model_pretrained))
elif self.backbone_pretrained is not None:
if os.path.exists(self.backbone_pretrained):
utils.load_pretrained_model(self.backbone,
self.backbone_pretrained)
else:
raise Exception('Pretrained model is not found: {}'.format(
self.backbone_pretrained))
else:
logger.warning('No pretrained model to load, train from scratch')
class ConvBNLayer(nn.Layer):
def __init__(self,
in_channels,
out_channels,
kernel_size,
stride=1,
groups=1,
act="relu"):
super(ConvBNLayer, self).__init__()
self._conv = Conv2d(
in_channels=in_channels,
out_channels=out_channels,
kernel_size=kernel_size,
stride=stride,
padding=(kernel_size - 1) // 2,
groups=groups,
bias_attr=False)
self._batch_norm = BatchNorm(out_channels)
self.act = activation.Activation(act=act)
def forward(self, input):
y = self._conv(input)
y = self._batch_norm(y)
y = self.act(y)
return y
@manager.MODELS.add_component @manager.MODELS.add_component
def fcn_hrnet_w18_small_v1(*args, **kwargs): def fcn_hrnet_w18_small_v1(*args, **kwargs):
......
dygraph/paddleseg/models/unet.py
浏览文件 @ 8f75ee50
...@@ -14,83 +14,47 @@ ...@@ -14,83 +14,47 @@
import os import os
import paddle.fluid as fluid import paddle
from paddle.fluid.dygraph import Conv2D, Pool2D import paddle.nn as nn
import paddle.nn.functional as F
from paddle.nn import Conv2d
from paddle.nn import SyncBatchNorm as BatchNorm from paddle.nn import SyncBatchNorm as BatchNorm
from paddleseg.cvlibs import manager from paddleseg.cvlibs import manager
from paddleseg import utils from paddleseg import utils
from paddleseg.models.common import layer_libs
class UNet(fluid.dygraph.Layer): @manager.MODELS.add_component
class UNet(nn.Layer):
""" """
U-Net: Convolutional Networks for Biomedical Image Segmentation. U-Net: Convolutional Networks for Biomedical Image Segmentation.
https://arxiv.org/abs/1505.04597 https://arxiv.org/abs/1505.04597
Args: Args:
num_classes (int): the unique number of target classes. num_classes (int): the unique number of target classes.
pretrained_model (str): the path of pretrained model. pretrained (str): the path of pretrained model for fine tuning.
ignore_index (int): the value of ground-truth mask would be ignored while computing loss or doing evaluation. Default 255.
""" """
def __init__(self, num_classes, model_pretrained=None, ignore_index=255): def __init__(self, num_classes, pretrained=None):
super(UNet, self).__init__() super(UNet, self).__init__()
self.encode = UnetEncoder() self.encode = UnetEncoder()
self.decode = UnetDecode() self.decode = UnetDecode()
self.get_logit = GetLogit(64, num_classes) self.get_logit = GetLogit(64, num_classes)
self.ignore_index = ignore_index
self.EPS = 1e-5
self.init_weight(model_pretrained) utils.load_entire_model(self, pretrained)
def forward(self, x, label=None): def forward(self, x, label=None):
logit_list = []
encode_data, short_cuts = self.encode(x) encode_data, short_cuts = self.encode(x)
decode_data = self.decode(encode_data, short_cuts) decode_data = self.decode(encode_data, short_cuts)
logit = self.get_logit(decode_data) logit = self.get_logit(decode_data)
if self.training: logit_list.append(logit)
return self._get_loss(logit, label) return logit_list
else:
score_map = fluid.layers.softmax(logit, axis=1)
score_map = fluid.layers.transpose(score_map, [0, 2, 3, 1]) class UnetEncoder(nn.Layer):
pred = fluid.layers.argmax(score_map, axis=3)
pred = fluid.layers.unsqueeze(pred, axes=[3])
return pred, score_map
def init_weight(self, pretrained_model=None):
"""
Initialize the parameters of model parts.
Args:
pretrained_model ([str], optional): the path of pretrained model. Defaults to None.
"""
if pretrained_model is not None:
if os.path.exists(pretrained_model):
utils.load_pretrained_model(self, pretrained_model)
else:
raise Exception('Pretrained model is not found: {}'.format(
pretrained_model))
def _get_loss(self, logit, label):
logit = fluid.layers.transpose(logit, [0, 2, 3, 1])
label = fluid.layers.transpose(label, [0, 2, 3, 1])
mask = label != self.ignore_index
mask = fluid.layers.cast(mask, 'float32')
loss, probs = fluid.layers.softmax_with_cross_entropy(
logit,
label,
ignore_index=self.ignore_index,
return_softmax=True,
axis=-1)
loss = loss * mask
avg_loss = fluid.layers.mean(loss) / (
fluid.layers.mean(mask) + self.EPS)
label.stop_gradient = True
mask.stop_gradient = True
return avg_loss
class UnetEncoder(fluid.dygraph.Layer):
def __init__(self): def __init__(self):
super(UnetEncoder, self).__init__() super(UnetEncoder, self).__init__()
self.double_conv = DoubleConv(3, 64) self.double_conv = DoubleConv(3, 64)
...@@ -113,7 +77,7 @@ class UnetEncoder(fluid.dygraph.Layer): ...@@ -113,7 +77,7 @@ class UnetEncoder(fluid.dygraph.Layer):
return x, short_cuts return x, short_cuts
class UnetDecode(fluid.dygraph.Layer): class UnetDecode(nn.Layer):
def __init__(self): def __init__(self):
super(UnetDecode, self).__init__() super(UnetDecode, self).__init__()
self.up1 = Up(512, 256) self.up1 = Up(512, 256)
...@@ -129,20 +93,20 @@ class UnetDecode(fluid.dygraph.Layer): ...@@ -129,20 +93,20 @@ class UnetDecode(fluid.dygraph.Layer):
return x return x
class DoubleConv(fluid.dygraph.Layer): class DoubleConv(nn.Layer):
def __init__(self, num_channels, num_filters): def __init__(self, num_channels, num_filters):
super(DoubleConv, self).__init__() super(DoubleConv, self).__init__()
self.conv0 = Conv2D( self.conv0 = Conv2d(
num_channels=num_channels, in_channels=num_channels,
num_filters=num_filters, out_channels=num_filters,
filter_size=3, kernel_size=3,
stride=1, stride=1,
padding=1) padding=1)
self.bn0 = BatchNorm(num_filters) self.bn0 = BatchNorm(num_filters)
self.conv1 = Conv2D( self.conv1 = Conv2d(
num_channels=num_filters, in_channels=num_filters,
num_filters=num_filters, out_channels=num_filters,
filter_size=3, kernel_size=3,
stride=1, stride=1,
padding=1) padding=1)
self.bn1 = BatchNorm(num_filters) self.bn1 = BatchNorm(num_filters)
...@@ -150,18 +114,17 @@ class DoubleConv(fluid.dygraph.Layer): ...@@ -150,18 +114,17 @@ class DoubleConv(fluid.dygraph.Layer):
def forward(self, x): def forward(self, x):
x = self.conv0(x) x = self.conv0(x)
x = self.bn0(x) x = self.bn0(x)
x = fluid.layers.relu(x) x = F.relu(x)
x = self.conv1(x) x = self.conv1(x)
x = self.bn1(x) x = self.bn1(x)
x = fluid.layers.relu(x) x = F.relu(x)
return x return x
class Down(fluid.dygraph.Layer): class Down(nn.Layer):
def __init__(self, num_channels, num_filters): def __init__(self, num_channels, num_filters):
super(Down, self).__init__() super(Down, self).__init__()
self.max_pool = Pool2D( self.max_pool = nn.MaxPool2d(kernel_size=2, stride=2)
pool_size=2, pool_type='max', pool_stride=2, pool_padding=0)
self.double_conv = DoubleConv(num_channels, num_filters) self.double_conv = DoubleConv(num_channels, num_filters)
def forward(self, x): def forward(self, x):
...@@ -170,34 +133,28 @@ class Down(fluid.dygraph.Layer): ...@@ -170,34 +133,28 @@ class Down(fluid.dygraph.Layer):
return x return x
class Up(fluid.dygraph.Layer): class Up(nn.Layer):
def __init__(self, num_channels, num_filters): def __init__(self, num_channels, num_filters):
super(Up, self).__init__() super(Up, self).__init__()
self.double_conv = DoubleConv(2 * num_channels, num_filters) self.double_conv = DoubleConv(2 * num_channels, num_filters)
def forward(self, x, short_cut): def forward(self, x, short_cut):
short_cut_shape = fluid.layers.shape(short_cut) x = F.resize_bilinear(x, short_cut.shape[2:])
x = fluid.layers.resize_bilinear(x, short_cut_shape[2:]) x = paddle.concat([x, short_cut], axis=1)
x = fluid.layers.concat([x, short_cut], axis=1)
x = self.double_conv(x) x = self.double_conv(x)
return x return x
class GetLogit(fluid.dygraph.Layer): class GetLogit(nn.Layer):
def __init__(self, num_channels, num_classes): def __init__(self, num_channels, num_classes):
super(GetLogit, self).__init__() super(GetLogit, self).__init__()
self.conv = Conv2D( self.conv = Conv2d(
num_channels=num_channels, in_channels=num_channels,
num_filters=num_classes, out_channels=num_classes,
filter_size=3, kernel_size=3,
stride=1, stride=1,
padding=1) padding=1)
def forward(self, x): def forward(self, x):
x = self.conv(x) x = self.conv(x)
return x return x
@manager.MODELS.add_component
def unet(*args, **kwargs):
return UNet(*args, **kwargs)
dygraph/train.py
浏览文件 @ 8f75ee50
...@@ -87,7 +87,7 @@ def parse_args(): ...@@ -87,7 +87,7 @@ def parse_args():
def main(args): def main(args):
env_info = get_environ_info() env_info = get_environ_info()
info = ['{}: {}'.format(k, v) for k, v in env_info.items()] info = ['{}: {}'.format(k, v) for k, v in env_info.items()]
info = '\n'.join(['\n', format('Environment Information', '-^48s')] + info + info = '\n'.join(['', format('Environment Information', '-^48s')] + info +
['-' * 48]) ['-' * 48])
logger.info(info) logger.info(info)
......
Markdown is supported
0% 或 .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑！
想要评论请 注册 或