Source code for gluoncv.model_zoo.deeplabv3b_plus
"""DeepLabV3+ with wideresnet backbone for semantic segmentation"""
# pylint: disable=missing-docstring,arguments-differ,unused-argument
from mxnet.gluon import nn
from mxnet.context import cpu
from mxnet.gluon.nn import HybridBlock
from .wideresnet import wider_resnet38_a2
__all__ = ['DeepLabWV3Plus', 'get_deeplabv3b_plus', 'get_deeplab_v3b_plus_wideresnet_citys']
[docs]class DeepLabWV3Plus(HybridBlock):
r"""DeepLabWV3Plus
Parameters
----------
nclass : int
Number of categories for the training dataset.
backbone : string
Pre-trained dilated backbone network type (default:'wideresnet').
norm_layer : object
Normalization layer used in backbone network (default: :class:`mxnet.gluon.nn.BatchNorm`;
for Synchronized Cross-GPU BachNormalization).
aux : bool
Auxiliary loss.
Reference:
Chen, Liang-Chieh, et al. "Encoder-Decoder with Atrous Separable Convolution for Semantic
Image Segmentation.", https://arxiv.org/abs/1802.02611, ECCV 2018
"""
def __init__(self, nclass, backbone='wideresnet', aux=False, ctx=cpu(), pretrained_base=True,
height=None, width=None, base_size=520, crop_size=480, dilated=True, **kwargs):
super(DeepLabWV3Plus, self).__init__()
height = height if height is not None else crop_size
width = width if width is not None else crop_size
self._up_kwargs = {'height': height, 'width': width}
self.base_size = base_size
self.crop_size = crop_size
#print('self.crop_size', self.crop_size)
kwargs.pop('root', None)
with self.name_scope():
pretrained = wider_resnet38_a2(classes=1000, dilation=True)
pretrained.initialize(ctx=ctx)
self.mod1 = pretrained.mod1
self.mod2 = pretrained.mod2
self.mod3 = pretrained.mod3
self.mod4 = pretrained.mod4
self.mod5 = pretrained.mod5
self.mod6 = pretrained.mod6
self.mod7 = pretrained.mod7
self.pool2 = pretrained.pool2
self.pool3 = pretrained.pool3
del pretrained
self.head = _DeepLabHead(nclass, height=height//2, width=width//2, **kwargs)
self.head.initialize(ctx=ctx)
[docs] def hybrid_forward(self, F, x):
outputs = []
x = self.mod1(x)
m2 = self.mod2(self.pool2(x))
x = self.mod3(self.pool3(m2))
x = self.mod4(x)
x = self.mod5(x)
x = self.mod6(x)
x = self.mod7(x)
x = self.head(x, m2)
x = F.contrib.BilinearResize2D(x, **self._up_kwargs)
outputs.append(x)
return tuple(outputs)
def demo(self, x):
return self.predict(x)
def evaluate(self, x):
return self.forward(x)[0]
def predict(self, x):
h, w = x.shape[2:]
self._up_kwargs['height'] = h
self._up_kwargs['width'] = w
x = self.mod1(x)
m2 = self.mod2(self.pool2(x))
x = self.mod3(self.pool3(m2))
x = self.mod4(x)
x = self.mod5(x)
x = self.mod6(x)
x = self.mod7(x)
x = self.head.demo(x, m2)
import mxnet.ndarray as F
x = F.contrib.BilinearResize2D(x, **self._up_kwargs)
return x
class _DeepLabHead(HybridBlock):
def __init__(self, nclass, c1_channels=128, norm_layer=nn.BatchNorm, norm_kwargs=None,
height=240, width=240, **kwargs):
super(_DeepLabHead, self).__init__()
self._up_kwargs = {'height': height, 'width': width}
with self.name_scope():
self.aspp = _ASPP(in_channels=4096, atrous_rates=[12, 24, 36], norm_layer=norm_layer,
norm_kwargs=norm_kwargs, height=height//4, width=width//4, **kwargs)
self.c1_block = nn.HybridSequential(prefix='bot_fine_')
self.c1_block.add(nn.Conv2D(in_channels=c1_channels, channels=48,
kernel_size=1, use_bias=False))
self.block = nn.HybridSequential(prefix='final_')
self.block.add(nn.Conv2D(in_channels=304, channels=256,
kernel_size=3, padding=1, use_bias=False))
self.block.add(norm_layer(in_channels=256,
**({} if norm_kwargs is None else norm_kwargs)))
self.block.add(nn.Activation('relu'))
self.block.add(nn.Conv2D(in_channels=256, channels=256,
kernel_size=3, padding=1, use_bias=False))
self.block.add(norm_layer(in_channels=256,
**({} if norm_kwargs is None else norm_kwargs)))
self.block.add(nn.Activation('relu'))
self.block.add(nn.Conv2D(in_channels=256, channels=nclass,
kernel_size=1, use_bias=False))
def hybrid_forward(self, F, x, c1):
c1 = self.c1_block(c1)
x = self.aspp(x)
x = F.contrib.BilinearResize2D(x, **self._up_kwargs)
return self.block(F.concat(c1, x, dim=1))
def demo(self, x, c1):
h, w = c1.shape[2:]
self._up_kwargs['height'] = h
self._up_kwargs['width'] = w
c1 = self.c1_block(c1)
x = self.aspp.demo(x)
import mxnet.ndarray as F
x = F.contrib.BilinearResize2D(x, **self._up_kwargs)
return self.block(F.concat(c1, x, dim=1))
def _ASPPConv(in_channels, out_channels, atrous_rate, norm_layer, norm_kwargs):
block = nn.HybridSequential()
with block.name_scope():
block.add(nn.Conv2D(in_channels=in_channels, channels=out_channels,
kernel_size=3, padding=atrous_rate,
dilation=atrous_rate, use_bias=False))
block.add(norm_layer(in_channels=out_channels,
**({} if norm_kwargs is None else norm_kwargs)))
block.add(nn.Activation('relu'))
return block
class _AsppPooling(nn.HybridBlock):
def __init__(self, in_channels, out_channels, norm_layer, norm_kwargs,
height=60, width=60, **kwargs):
super(_AsppPooling, self).__init__()
self.gap = nn.HybridSequential()
self._up_kwargs = {'height': height, 'width': width}
with self.gap.name_scope():
self.gap.add(nn.GlobalAvgPool2D())
self.gap.add(nn.Conv2D(in_channels=in_channels, channels=out_channels,
kernel_size=1, use_bias=False))
self.gap.add(norm_layer(in_channels=out_channels,
**({} if norm_kwargs is None else norm_kwargs)))
self.gap.add(nn.Activation("relu"))
def hybrid_forward(self, F, x):
pool = self.gap(x)
return F.contrib.BilinearResize2D(pool, **self._up_kwargs)
def demo(self, x):
h, w = x.shape[2:]
self._up_kwargs['height'] = h
self._up_kwargs['width'] = w
pool = self.gap(x)
import mxnet.ndarray as F
return F.contrib.BilinearResize2D(pool, **self._up_kwargs)
class _ASPP(nn.HybridBlock):
def __init__(self, in_channels, atrous_rates, norm_layer, norm_kwargs,
height=60, width=60):
super(_ASPP, self).__init__()
out_channels = 256
self.b0 = nn.HybridSequential()
self.b0.add(nn.Conv2D(in_channels=in_channels, channels=out_channels,
kernel_size=1, use_bias=False))
self.b0.add(norm_layer(in_channels=out_channels,
**({} if norm_kwargs is None else norm_kwargs)))
self.b0.add(nn.Activation("relu"))
rate1, rate2, rate3 = tuple(atrous_rates)
self.b1 = _ASPPConv(in_channels, out_channels, rate1, norm_layer, norm_kwargs)
self.b2 = _ASPPConv(in_channels, out_channels, rate2, norm_layer, norm_kwargs)
self.b3 = _ASPPConv(in_channels, out_channels, rate3, norm_layer, norm_kwargs)
self.b4 = _AsppPooling(in_channels, out_channels, norm_layer=norm_layer,
norm_kwargs=norm_kwargs, height=height, width=width)
self.project = nn.HybridSequential(prefix='bot_aspp_')
self.project.add(nn.Conv2D(in_channels=5*out_channels, channels=out_channels,
kernel_size=1, use_bias=False))
def hybrid_forward(self, F, x):
feat1 = self.b0(x)
feat2 = self.b1(x)
feat3 = self.b2(x)
feat4 = self.b3(x)
x = self.b4(x)
x = F.concat(x, feat1, feat2, feat3, feat4, dim=1)
return self.project(x)
def demo(self, x):
feat1 = self.b0(x)
feat2 = self.b1(x)
feat3 = self.b2(x)
feat4 = self.b3(x)
x = self.b4.demo(x)
import mxnet.ndarray as F
x = F.concat(x, feat1, feat2, feat3, feat4, dim=1)
return self.project(x)
[docs]def get_deeplabv3b_plus(dataset='citys', backbone='wideresnet', pretrained=False,
root='~/.mxnet/models', ctx=cpu(0), **kwargs):
r"""DeepLabWV3Plus
Parameters
----------
dataset : str, default pascal_voc
The dataset that model pretrained on. (pascal_voc, ade20k, citys)
pretrained : bool or str
Boolean value controls whether to load the default pretrained weights for model.
String value represents the hashtag for a certain version of pretrained weights.
ctx : Context, default CPU
The context in which to load the pretrained weights.
root : str, default '~/.mxnet/models'
Location for keeping the model parameters.
Examples
--------
>>> model = get_deeplabv3b_plus(dataset='citys', backbone='wideresnet', pretrained=False)
>>> print(model)
"""
acronyms = {
'pascal_voc': 'voc',
'pascal_aug': 'voc',
'ade20k': 'ade',
'coco': 'coco',
'citys': 'citys',
}
from ..data import datasets
# infer number of classes
if pretrained:
kwargs['pretrained_base'] = False
kwargs['root'] = root
model = DeepLabWV3Plus(datasets[dataset].NUM_CLASS, backbone=backbone, ctx=ctx, **kwargs)
model.classes = datasets[dataset].CLASSES
if pretrained:
from .model_store import get_model_file
model.load_parameters(get_model_file('deeplab_v3b_plus_%s_%s'%(backbone, acronyms[dataset]),
tag=pretrained, root=root), ctx=ctx)
return model
[docs]def get_deeplab_v3b_plus_wideresnet_citys(**kwargs):
r"""DeepLabWV3Plus
Parameters
----------
pretrained : bool or str
Boolean value controls whether to load the default pretrained weights for model.
String value represents the hashtag for a certain version of pretrained weights.
ctx : Context, default CPU
The context in which to load the pretrained weights.
root : str, default '~/.mxnet/models'
Location for keeping the model parameters.
Examples
--------
>>> model = get_deeplab_v3b_plus_wideresnet_citys(pretrained=True)
>>> print(model)
"""
return get_deeplabv3b_plus('citys', 'wideresnet', **kwargs)