gy15635282365
/
tensorlayer3

#!/usr/bin/env python
# -*- coding: utf-8 -*-

import os
import unittest

os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'

import tensorflow as tf
import tensorlayer as tl

from tests.utils import CustomTestCase


class Layer_Convolution_1D_Test(CustomTestCase):

    @classmethod
    def setUpClass(cls):

        def get_network_1d(inputs, reuse=False):

            with tf.variable_scope("1D_network", reuse=reuse):
                net = tl.layers.InputLayer(inputs)

                net1 = tl.layers.Conv1d(net, name="Conv1d")  # 2 params
                net2 = tl.layers.SeparableConv1d(net1, name="SeparableConv1d")  # 3 params
                net3 = tl.layers.MaxPool1d(net2, (1, ), name="MaxPool1d")  # 0 params
                net4 = tl.layers.MeanPool1d(net3, (1, ), name="MeanPool1d")  # 0 params

                # HAO Test
                net5 = tl.layers.Conv1d(net4, name="Conv1d1")  # 2 params
                net6 = tl.layers.SeparableConv1d(net5, name="SeparableConv1d1")  # 3 params
                net7 = tl.layers.SeparableConv1d(net6, name="SeparableConv1d2")  # 3 params

            return [net, net1, net2, net3, net4, net5, net6, net7]

        input_pl_train = tf.placeholder(tf.float32, [None, 32, 3])
        input_plh_test = tf.placeholder(tf.float32, [None, 32, 3])

        cls.network_1 = get_network_1d(input_pl_train, reuse=False)
        cls.network_2 = get_network_1d(input_plh_test, reuse=True)

    @classmethod
    def tearDownClass(cls):
        tf.reset_default_graph()

    def test_layer_net0(self):
        self.assertEqual(len(self.network_1[0].all_params), 0)
        self.assertEqual(len(self.network_2[0].all_params), 0)

    def test_layer_net1(self):
        self.assertEqual(len(self.network_1[1].all_params), 2)
        self.assertEqual(len(self.network_2[1].all_params), 2)

    def test_layer_net2(self):
        self.assertEqual(len(self.network_1[2].all_params), 5)
        self.assertEqual(len(self.network_2[2].all_params), 5)

    def test_layer_net3(self):
        self.assertEqual(len(self.network_1[3].all_params), 5)
        self.assertEqual(len(self.network_2[3].all_params), 5)

    def test_layer_net4(self):
        self.assertEqual(len(self.network_1[4].all_params), 5)
        self.assertEqual(len(self.network_2[4].all_params), 5)

    def test_layer_net5(self):
        self.assertEqual(len(self.network_1[5].all_params), 7)
        self.assertEqual(len(self.network_2[5].all_params), 7)

    def test_layer_net6(self):
        self.assertEqual(len(self.network_1[6].all_params), 10)
        self.assertEqual(len(self.network_2[6].all_params), 10)

    def test_layer_net7(self):
        self.assertEqual(len(self.network_1[7].all_params), 13)
        self.assertEqual(len(self.network_2[7].all_params), 13)


class Layer_Convolution_2D_Test(CustomTestCase):

    @classmethod
    def setUpClass(cls):

        def get_network_2d(inputs, reuse=False):

            with tf.variable_scope("2D_network", reuse=reuse):
                net = tl.layers.InputLayer(inputs)

                net1 = tl.layers.Conv2d(net, name="Conv2d")  # 2 params
                net2 = tl.layers.DeConv2d(net1, name="DeConv2d")  # 2 params
                net3 = tl.layers.SeparableConv2d(net2, name="SeparableConv2d")  # 3 params
                net4 = tl.layers.MaxPool2d(net3, (1, 1), name="MaxPool2d")  # 0 params
                net5 = tl.layers.MeanPool2d(net4, (1, 1), name="MeanPool2d")  # 0 params

                # HAO Test
                net6 = tl.layers.Conv2d(net5, name="Conv2d1")  # 2 params
                net7 = tl.layers.DeConv2d(net6, name="DeConv2d1")  # 2 params
                net8 = tl.layers.DeConv2d(net7, name="DeConv2d2")  # 2 params
                net9 = tl.layers.SeparableConv2d(net8, name="SeparableConv2d1")  # 3 params

            return [net, net1, net2, net3, net4, net5, net6, net7, net8, net9]

        input_pl_train = tf.placeholder(tf.float32, [None, 32, 32, 3])
        input_plh_test = tf.placeholder(tf.float32, [None, 32, 32, 3])

        cls.network_1 = get_network_2d(input_pl_train, reuse=False)
        cls.network_2 = get_network_2d(input_plh_test, reuse=True)

    def test_layer_net0(self):
        self.assertEqual(len(self.network_1[0].all_params), 0)
        self.assertEqual(len(self.network_2[0].all_params), 0)

    def test_layer_net1(self):
        self.assertEqual(len(self.network_1[1].all_params), 2)
        self.assertEqual(len(self.network_2[1].all_params), 2)

    def test_layer_net2(self):
        self.assertEqual(len(self.network_1[2].all_params), 4)
        self.assertEqual(len(self.network_2[2].all_params), 4)

    def test_layer_net3(self):
        self.assertEqual(len(self.network_1[3].all_params), 7)
        self.assertEqual(len(self.network_2[3].all_params), 7)

    def test_layer_net4(self):
        self.assertEqual(len(self.network_1[4].all_params), 7)
        self.assertEqual(len(self.network_2[4].all_params), 7)

    def test_layer_net5(self):
        self.assertEqual(len(self.network_1[5].all_params), 7)
        self.assertEqual(len(self.network_2[5].all_params), 7)

    def test_layer_net6(self):
        self.assertEqual(len(self.network_1[6].all_params), 9)
        self.assertEqual(len(self.network_2[6].all_params), 9)

    def test_layer_net7(self):
        self.assertEqual(len(self.network_1[7].all_params), 11)
        self.assertEqual(len(self.network_2[7].all_params), 11)

    def test_layer_net8(self):
        self.assertEqual(len(self.network_1[8].all_params), 13)
        self.assertEqual(len(self.network_2[8].all_params), 13)

    def test_layer_net9(self):
        self.assertEqual(len(self.network_1[9].all_params), 16)
        self.assertEqual(len(self.network_2[9].all_params), 16)

    @classmethod
    def tearDownClass(cls):
        tf.reset_default_graph()


class Layer_Convolution_3D_Test(CustomTestCase):

    @classmethod
    def setUpClass(cls):

        def get_network_3d(inputs, reuse=False):

            with tf.variable_scope("3D_network", reuse=reuse):
                net = tl.layers.InputLayer(inputs)

                net1 = tl.layers.Conv3dLayer(
                    net, shape=(2, 2, 2, 3, 32), strides=(1, 2, 2, 2, 1), name="Conv3dLayer"
                )  # 2 params
                net2 = tl.layers.DeConv3d(net1, name="DeConv3d")  # 2 params
                net3 = tl.layers.MaxPool3d(net2, (1, 1, 1), name="MaxPool3d")  # 0 params
                net4 = tl.layers.MeanPool3d(net3, (1, 1, 1), name="MeanPool3d")  # 0 params

                # HAO Test
                net5 = tl.layers.DeConv3d(net4, name="DeConv3d1")  # 2 params

                return [net, net1, net2, net3, net4, net5]

        input_pl_train = tf.placeholder(tf.float32, [None, 32, 32, 32, 3])
        input_plh_test = tf.placeholder(tf.float32, [None, 32, 32, 32, 3])

        cls.network_1 = get_network_3d(input_pl_train, reuse=False)
        cls.network_2 = get_network_3d(input_plh_test, reuse=True)

    @classmethod
    def tearDownClass(cls):
        tf.reset_default_graph()

    def test_layer_net0(self):
        self.assertEqual(len(self.network_1[0].all_params), 0)
        self.assertEqual(len(self.network_2[0].all_params), 0)

    def test_layer_net1(self):
        self.assertEqual(len(self.network_1[1].all_params), 2)
        self.assertEqual(len(self.network_2[1].all_params), 2)

    def test_layer_net2(self):
        self.assertEqual(len(self.network_1[2].all_params), 4)
        self.assertEqual(len(self.network_2[2].all_params), 4)

    def test_layer_net3(self):
        self.assertEqual(len(self.network_1[3].all_params), 4)
        self.assertEqual(len(self.network_2[3].all_params), 4)

    def test_layer_net4(self):
        self.assertEqual(len(self.network_1[4].all_params), 4)
        self.assertEqual(len(self.network_2[4].all_params), 4)

    def test_layer_net5(self):
        self.assertEqual(len(self.network_1[5].all_params), 6)
        self.assertEqual(len(self.network_2[5].all_params), 6)


if __name__ == '__main__':

    tf.logging.set_verbosity(tf.logging.DEBUG)
    tl.logging.set_verbosity(tl.logging.DEBUG)

    unittest.main()