resnet.py

from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import paddle
import paddle.fluid as fluid
import math
from paddle.fluid.param_attr import ParamAttr

__all__ = ["ResNet", "ResNet34", "ResNet50", "ResNet101", "ResNet152"]

train_parameters = {
    "input_size": [3, 224, 224],
    "input_mean": [0.485, 0.456, 0.406],
    "input_std": [0.229, 0.224, 0.225],
    "learning_strategy": {
        "name": "piecewise_decay",
        "batch_size": 256,
        "epochs": [10, 16, 30],
        "steps": [0.1, 0.01, 0.001, 0.0001]
    }
}


class ResNet():
    def __init__(self, layers=50, prefix_name=''):
        self.params = train_parameters
        self.layers = layers
        self.prefix_name = prefix_name

    def net(self, input, class_dim=1000, conv1_name='conv1', fc_name=None):
        layers = self.layers
        prefix_name = self.prefix_name if self.prefix_name is '' else self.prefix_name + '_'
        supported_layers = [34, 50, 101, 152]
        assert layers in supported_layers, \
            "supported layers are {} but input layer is {}".format(supported_layers, layers)

        if layers == 34 or layers == 50:
            depth = [3, 4, 6, 3]
        elif layers == 101:
            depth = [3, 4, 23, 3]
        elif layers == 152:
            depth = [3, 8, 36, 3]
        num_filters = [64, 128, 256, 512]

        # TODO(wanghaoshuang@baidu.com):
        # fix name("conv1") conflict between student and teacher in distillation.
        conv = self.conv_bn_layer(
            input=input,
            num_filters=64,
            filter_size=7,
            stride=2,
            act='relu',
            name=prefix_name + conv1_name)
        conv = fluid.layers.pool2d(
            input=conv,
            pool_size=3,
            pool_stride=2,
            pool_padding=1,
            pool_type='max')

        if layers >= 50:
            for block in range(len(depth)):
                for i in range(depth[block]):
                    if layers in [101, 152] and block == 2:
                        if i == 0:
                            conv_name = "res" + str(block + 2) + "a"
                        else:
                            conv_name = "res" + str(block + 2) + "b" + str(i)
                    else:
                        conv_name = "res" + str(block + 2) + chr(97 + i)
                    conv_name = prefix_name + conv_name
                    conv = self.bottleneck_block(
                        input=conv,
                        num_filters=num_filters[block],
                        stride=2 if i == 0 and block != 0 else 1,
                        name=conv_name)

            pool = fluid.layers.pool2d(
                input=conv, pool_size=7, pool_type='avg', global_pooling=True)
            stdv = 1.0 / math.sqrt(pool.shape[1] * 1.0)
            fc_name = fc_name if fc_name is None else prefix_name + fc_name
            out = fluid.layers.fc(input=pool,
                                  size=class_dim,
                                  act='softmax',
                                  name=fc_name,
                                  param_attr=fluid.param_attr.ParamAttr(
                                      initializer=fluid.initializer.Uniform(
                                          -stdv, stdv)))
        else:
            for block in range(len(depth)):
                for i in range(depth[block]):
                    conv_name = "res" + str(block + 2) + chr(97 + i)
                    conv_name = prefix_name + conv_name
                    conv = self.basic_block(
                        input=conv,
                        num_filters=num_filters[block],
                        stride=2 if i == 0 and block != 0 else 1,
                        is_first=block == i == 0,
                        name=conv_name)

            pool = fluid.layers.pool2d(
                input=conv, pool_type='avg', global_pooling=True)
            stdv = 1.0 / math.sqrt(pool.shape[1] * 1.0)
            fc_name = fc_name if fc_name is None else prefix_name + fc_name
            out = fluid.layers.fc(
                input=pool,
                size=class_dim,
                act='softmax',
                name=fc_name,
                param_attr=fluid.param_attr.ParamAttr(
                    initializer=fluid.initializer.Uniform(-stdv, stdv)))

        return out

    def conv_bn_layer(self,
                      input,
                      num_filters,
                      filter_size,
                      stride=1,
                      groups=1,
                      act=None,
                      name=None):
        conv = fluid.layers.conv2d(
            input=input,
            num_filters=num_filters,
            filter_size=filter_size,
            stride=stride,
            padding=(filter_size - 1) // 2,
            groups=groups,
            act=None,
            param_attr=ParamAttr(name=name + "_weights"),
            bias_attr=False,
            name=name + '.conv2d.output.1')
        if self.prefix_name == '':
            if name == "conv1":
                bn_name = "bn_" + name
            else:
                bn_name = "bn" + name[3:]
        else:
            if name.split("_")[1] == "conv1":
                bn_name = name.split("_", 1)[0] + "_bn_" + name.split("_",
                                                                      1)[1]
            else:
                bn_name = name.split("_", 1)[0] + "_bn" + name.split("_",
                                                                     1)[1][3:]
        return fluid.layers.batch_norm(
            input=conv,
            act=act,
            name=bn_name + '.output.1',
            param_attr=ParamAttr(name=bn_name + '_scale'),
            bias_attr=ParamAttr(bn_name + '_offset'),
            moving_mean_name=bn_name + '_mean',
            moving_variance_name=bn_name + '_variance', )

    def shortcut(self, input, ch_out, stride, is_first, name):
        ch_in = input.shape[1]
        if ch_in != ch_out or stride != 1 or is_first == True:
            return self.conv_bn_layer(input, ch_out, 1, stride, name=name)
        else:
            return input

    def bottleneck_block(self, input, num_filters, stride, name):
        conv0 = self.conv_bn_layer(
            input=input,
            num_filters=num_filters,
            filter_size=1,
            act='relu',
            name=name + "_branch2a")
        conv1 = self.conv_bn_layer(
            input=conv0,
            num_filters=num_filters,
            filter_size=3,
            stride=stride,
            act='relu',
            name=name + "_branch2b")
        conv2 = self.conv_bn_layer(
            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")

        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):
        conv0 = self.conv_bn_layer(
            input=input,
            num_filters=num_filters,
            filter_size=3,
            act='relu',
            stride=stride,
            name=name + "_branch2a")
        conv1 = self.conv_bn_layer(
            input=conv0,
            num_filters=num_filters,
            filter_size=3,
            act=None,
            name=name + "_branch2b")
        short = self.shortcut(
            input, num_filters, stride, is_first, name=name + "_branch1")
        return fluid.layers.elementwise_add(x=short, y=conv1, act='relu')


def ResNet34(prefix_name=''):
    model = ResNet(layers=34, prefix_name=prefix_name)
    return model


def ResNet50(prefix_name=''):
    model = ResNet(layers=50, prefix_name=prefix_name)
    return model


def ResNet101():
    model = ResNet(layers=101)
    return model


def ResNet152():
    model = ResNet(layers=152)
    return model