test_imperative_resnet.py

# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

import contextlib
import unittest
import numpy as np
import six

import paddle
import paddle.fluid as fluid
from paddle.fluid import core
from paddle.fluid.optimizer import SGDOptimizer
from paddle.fluid.imperative.nn import Conv2D, Pool2D, BatchNorm, FC
from paddle.fluid.imperative.base import to_variable
from test_imperative_base import new_program_scope

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": 1,
        "epochs": [30, 60, 90],
        "steps": [0.1, 0.01, 0.001, 0.0001]
    },
    "batch_size": 1,
    "lr": 0.1,
    "total_images": 1281164,
}


def optimizer_setting(params):
    ls = params["learning_strategy"]
    if ls["name"] == "piecewise_decay":
        if "total_images" not in params:
            total_images = 1281167
        else:
            total_images = params["total_images"]
        batch_size = ls["batch_size"]
        step = int(total_images / batch_size + 1)

        bd = [step * e for e in ls["epochs"]]
        base_lr = params["lr"]
        lr = []
        lr = [base_lr * (0.1**i) for i in range(len(bd) + 1)]
        optimizer = fluid.optimizer.SGD(learning_rate=params["lr"])
        #  optimizer = fluid.optimizer.Momentum(
    #  learning_rate=params["lr"],
    #  learning_rate=fluid.layers.piecewise_decay(
    #  boundaries=bd, values=lr),
    #  momentum=0.9,
    #  regularization=fluid.regularizer.L2Decay(1e-4))

    return optimizer


class ConvBNLayer(fluid.imperative.Layer):
    def __init__(self,
                 num_channels,
                 num_filters,
                 filter_size,
                 stride=1,
                 groups=1,
                 act=None):
        super(ConvBNLayer, self).__init__()

        self._conv = Conv2D(
            num_channels=num_channels,
            num_filters=num_filters,
            filter_size=filter_size,
            stride=stride,
            padding=(filter_size - 1) // 2,
            groups=groups,
            act=None,
            bias_attr=None)

        self._batch_norm = BatchNorm(num_filters, act=act)

    def forward(self, inputs):
        y = self._conv(inputs)
        y = self._batch_norm(y)

        return y


class BottleneckBlock(fluid.imperative.Layer):
    def __init__(self, num_channels, num_filters, stride, shortcut=True):
        super(BottleneckBlock, self).__init__()

        self.conv0 = ConvBNLayer(
            num_channels=num_channels,
            num_filters=num_filters,
            filter_size=1,
            act='relu')
        self.conv1 = ConvBNLayer(
            num_channels=num_filters,
            num_filters=num_filters,
            filter_size=3,
            stride=stride,
            act='relu')
        self.conv2 = ConvBNLayer(
            num_channels=num_filters,
            num_filters=num_filters * 4,
            filter_size=1,
            act=None)

        if not shortcut:
            self.short = ConvBNLayer(
                num_channels=num_channels,
                num_filters=num_filters * 4,
                filter_size=1,
                stride=stride)

        self.shortcut = shortcut

        self._num_channels_out = num_filters * 4

    def forward(self, inputs):
        y = self.conv0(inputs)
        conv1 = self.conv1(y)
        conv2 = self.conv2(conv1)

        if self.shortcut:
            short = inputs
        else:
            short = self.short(inputs)

        return fluid.layers.elementwise_add(x=short, y=conv2, act='relu')


class ResNet(fluid.imperative.Layer):
    def __init__(self, layers=50, class_dim=1000):
        super(ResNet, self).__init__()

        self.layers = layers
        supported_layers = [50, 101, 152]
        assert layers in supported_layers, \
            "supported layers are {} but input layer is {}".format(supported_layers, layers)

        if 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]

        self.conv = ConvBNLayer(
            num_channels=3, num_filters=64, filter_size=7, stride=2, act='relu')
        self.pool2d_max = Pool2D(
            pool_size=3, pool_stride=2, pool_padding=1, pool_type='max')

        self.bottleneck_block_list = []
        num_channels = 64
        for block in range(len(depth)):
            shortcut = False
            for i in range(depth[block]):
                bottleneck_block = BottleneckBlock(
                    num_channels=num_channels,
                    num_filters=num_filters[block],
                    stride=2 if i == 0 and block != 0 else 1,
                    shortcut=shortcut)
                num_channels = bottleneck_block._num_channels_out
                self.bottleneck_block_list.append(bottleneck_block)
                shortcut = True

        self.pool2d_avg = Pool2D(
            pool_size=7, pool_type='avg', global_pooling=True)

        import math
        stdv = 1.0 / math.sqrt(2048 * 1.0)

        self.out = FC(size=class_dim,
                      act='softmax',
                      param_attr=fluid.param_attr.ParamAttr(
                          initializer=fluid.initializer.Uniform(-stdv, stdv)))

    def forward(self, inputs):
        y = self.conv(inputs)
        y = self.pool2d_max(y)
        for bottleneck_block in self.bottleneck_block_list:
            y = bottleneck_block(y)
        y = self.pool2d_avg(y)
        y = self.out(y)
        return y


class TestImperativeResnet(unittest.TestCase):
    def test_resnet_gpu_float32(self):
        seed = 90

        batch_size = train_parameters["batch_size"]
        with fluid.imperative.guard():
            fluid.default_startup_program().random_seed = seed
            fluid.default_main_program().random_seed = seed

            resnet = ResNet()
            optimizer = optimizer_setting(train_parameters)
            train_reader = paddle.batch(
                paddle.dataset.flowers.train(), batch_size=batch_size)

            dy_param_init_value = {}
            for param in fluid.default_main_program().global_block(
            ).all_parameters():
                dy_param_init_value[param.name] = param._numpy()

            for batch_id, data in enumerate(train_reader()):
                if batch_id >= 1:
                    break

                x_data = np.array(
                    [x[0].reshape(3, 224, 224) for x in data]).astype('float32')
                y_data = np.array([x[1] for x in data]).astype('int64').reshape(
                    batch_size, 1)

                img = to_variable(x_data)
                label = to_variable(y_data)
                label._stop_gradient = True

                out = resnet(img)
                loss = fluid.layers.cross_entropy(input=out, label=label)
                avg_loss = fluid.layers.mean(x=loss)
                dy_out = avg_loss._numpy()

                if batch_id == 0:
                    for param in fluid.default_main_program().global_block(
                    ).all_parameters():
                        if param.name not in dy_param_init_value:
                            dy_param_init_value[param.name] = param._numpy()

                avg_loss._backward()
                optimizer.minimize(avg_loss)

                dy_param_value = {}
                for param in fluid.default_main_program().global_block(
                ).all_parameters():
                    dy_param_value[param.name] = param._numpy()

        with new_program_scope():
            fluid.default_startup_program().random_seed = seed
            fluid.default_main_program().random_seed = seed

            exe = fluid.Executor(fluid.CUDAPlace(0))

            resnet = ResNet()
            optimizer = optimizer_setting(train_parameters)
            train_reader = paddle.batch(
                paddle.dataset.flowers.train(), batch_size=batch_size)

            img = fluid.layers.data(
                name='pixel', shape=[3, 224, 224], dtype='float32')
            label = fluid.layers.data(name='label', shape=[1], dtype='int64')
            out = resnet(img)
            loss = fluid.layers.cross_entropy(input=out, label=label)
            avg_loss = fluid.layers.mean(x=loss)
            optimizer.minimize(avg_loss)

            # initialize params and fetch them
            static_param_init_value = {}
            static_param_name_list = []
            for param in fluid.default_startup_program().global_block(
            ).all_parameters():
                static_param_name_list.append(param.name)

            out = exe.run(fluid.default_startup_program(),
                          fetch_list=static_param_name_list)

            for i in range(len(static_param_name_list)):
                static_param_init_value[static_param_name_list[i]] = out[i]

            for batch_id, data in enumerate(train_reader()):
                if batch_id >= 1:
                    break

                x_data = np.array(
                    [x[0].reshape(3, 224, 224) for x in data]).astype('float32')
                y_data = np.array([x[1] for x in data]).astype('int64').reshape(
                    [batch_size, 1])

                fetch_list = [loss.name]
                fetch_list.extend(static_param_name_list)
                out = exe.run(fluid.default_main_program(),
                              feed={"pixel": x_data,
                                    "label": y_data},
                              fetch_list=fetch_list)

                static_param_value = {}
                static_out = out[0]
                for i in range(1, len(out)):
                    static_param_value[static_param_name_list[i - 1]] = out[i]

        self.assertTrue(np.allclose(static_out.all(), dy_out.all()))

        for key, value in six.iteritems(static_param_init_value):
            self.assertTrue(
                np.allclose(value.all(), dy_param_init_value[key].all()))

        for key, value in six.iteritems(static_param_value):
            if not np.allclose(value.all(), dy_param_value[key].all()):
                print(key)
                print(value, dy_param_value[key])
            self.assertTrue(np.allclose(value.all(), dy_param_value[key].all()))


if __name__ == '__main__':
    unittest.main()