• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1# Copyright 2020-2022 Huawei Technologies Co., Ltd
2#
3# Licensed under the Apache License, Version 2.0 (the "License");
4# you may not use this file except in compliance with the License.
5# You may obtain a copy of the License at
6#
7# http://www.apache.org/licenses/LICENSE-2.0
8#
9# Unless required by applicable law or agreed to in writing, software
10# distributed under the License is distributed on an "AS IS" BASIS,
11# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12# See the License for the specific language governing permissions and
13# limitations under the License.
14# ============================================================================
15"""dataset base and LeNet."""
16import os
17
18from mindspore import dataset as ds
19from mindspore.common import dtype as mstype
20import mindspore.dataset.transforms as C
21from mindspore.dataset.vision import Inter
22import mindspore.dataset.vision as CV
23from mindspore import nn, Tensor
24from mindspore.common.initializer import Normal
25from mindspore.ops import operations as P
26
27
28def create_mnist_dataset(mode='train', num_samples=2, batch_size=2):
29    """create dataset for train or test"""
30    mnist_path = '/home/workspace/mindspore_dataset/mnist'
31    num_parallel_workers = 1
32
33    # define dataset
34    mnist_ds = ds.MnistDataset(os.path.join(mnist_path, mode), num_samples=num_samples, shuffle=False)
35
36    resize_height, resize_width = 32, 32
37
38    # define map operations
39    resize_op = CV.Resize((resize_height, resize_width), interpolation=Inter.LINEAR)  # Bilinear mode
40    rescale_nml_op = CV.Rescale(1 / 0.3081, -1 * 0.1307 / 0.3081)
41    rescale_op = CV.Rescale(1.0 / 255.0, shift=0.0)
42    hwc2chw_op = CV.HWC2CHW()
43    type_cast_op = C.TypeCast(mstype.int32)
44
45    # apply map operations on images
46    mnist_ds = mnist_ds.map(operations=type_cast_op, input_columns="label", num_parallel_workers=num_parallel_workers)
47    mnist_ds = mnist_ds.map(operations=resize_op, input_columns="image", num_parallel_workers=num_parallel_workers)
48    mnist_ds = mnist_ds.map(operations=rescale_op, input_columns="image", num_parallel_workers=num_parallel_workers)
49    mnist_ds = mnist_ds.map(operations=rescale_nml_op, input_columns="image", num_parallel_workers=num_parallel_workers)
50    mnist_ds = mnist_ds.map(operations=hwc2chw_op, input_columns="image", num_parallel_workers=num_parallel_workers)
51
52    # apply DatasetOps
53    mnist_ds = mnist_ds.batch(batch_size=batch_size, drop_remainder=True)
54
55    return mnist_ds
56
57
58class LeNet5(nn.Cell):
59    """
60    Lenet network
61
62    Args:
63        num_class (int): Number of classes. Default: 10.
64        num_channel (int): Number of channels. Default: 1.
65
66    Returns:
67        Tensor, output tensor
68    Examples:
69        >>> LeNet(num_class=10)
70
71    """
72
73    def __init__(self, num_class=10, num_channel=1, include_top=True):
74        super(LeNet5, self).__init__()
75        self.conv1 = nn.Conv2d(num_channel, 6, 5, pad_mode='valid')
76        self.conv2 = nn.Conv2d(6, 16, 5, pad_mode='valid')
77        self.relu = nn.ReLU()
78        self.max_pool2d = nn.MaxPool2d(kernel_size=2, stride=2)
79        self.include_top = include_top
80        if self.include_top:
81            self.flatten = nn.Flatten()
82            self.fc1 = nn.Dense(16 * 5 * 5, 120, weight_init=Normal(0.02))
83            self.fc2 = nn.Dense(120, 84, weight_init=Normal(0.02))
84            self.fc3 = nn.Dense(84, num_class, weight_init=Normal(0.02))
85
86        self.scalar_summary = P.ScalarSummary()
87        self.image_summary = P.ImageSummary()
88        self.histogram_summary = P.HistogramSummary()
89        self.tensor_summary = P.TensorSummary()
90        self.channel = Tensor(num_channel)
91
92    def construct(self, x):
93        """construct."""
94        self.image_summary('image', x)
95        x = self.conv1(x)
96        self.histogram_summary('histogram', x)
97        x = self.relu(x)
98        self.tensor_summary('tensor', x)
99        x = self.relu(x)
100        x = self.max_pool2d(x)
101        self.scalar_summary('scalar', self.channel)
102        x = self.conv2(x)
103        x = self.relu(x)
104        x = self.max_pool2d(x)
105        if not self.include_top:
106            return x
107        x = self.flatten(x)
108        x = self.relu(self.fc1(x))
109        x = self.relu(self.fc2(x))
110        x = self.fc3(x)
111        return x
112