test_conv3d_op.py 7.7 KB
Newer Older
1
#   Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
D
dzhwinter 已提交
2
#
D
dzhwinter 已提交
3 4 5
# 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
D
dzhwinter 已提交
6
#
D
dzhwinter 已提交
7
#     http://www.apache.org/licenses/LICENSE-2.0
D
dzhwinter 已提交
8
#
D
dzhwinter 已提交
9 10 11 12 13 14
# 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.

C
chengduoZH 已提交
15 16
import unittest
import numpy as np
17

18
import paddle.fluid.core as core
C
chengduoZH 已提交
19 20 21
from op_test import OpTest


22 23 24 25 26 27 28
def conv3d_forward_naive(input, filter, group, conv_param):
    in_n, in_c, in_d, in_h, in_w = input.shape
    out_c, f_c, f_d, f_h, f_w = filter.shape
    assert f_c * group == in_c
    assert np.mod(out_c, group) == 0
    sub_out_c = out_c / group

C
chengduoZH 已提交
29 30 31 32 33 34 35
    stride, pad, dilation = conv_param['stride'], conv_param['pad'], conv_param[
        'dilations']

    out_d = 1 + (in_d + 2 * pad[0] - (dilation[0] * (f_d - 1) + 1)) / stride[0]
    out_h = 1 + (in_h + 2 * pad[1] - (dilation[1] * (f_h - 1) + 1)) / stride[1]
    out_w = 1 + (in_w + 2 * pad[2] - (dilation[2] * (f_w - 1) + 1)) / stride[2]

36 37
    out = np.zeros((in_n, out_c, out_d, out_h, out_w))

C
chengduoZH 已提交
38 39 40 41
    d_bolck_d = (dilation[0] * (f_d - 1) + 1)
    d_bolck_h = (dilation[1] * (f_h - 1) + 1)
    d_bolck_w = (dilation[2] * (f_w - 1) + 1)

42 43 44 45
    input_pad = np.pad(input, ((0, ), (0, ), (pad[0], ), (pad[1], ),
                               (pad[2], )),
                       mode='constant',
                       constant_values=0)
C
chengduoZH 已提交
46 47 48 49 50

    filter_dilation = np.zeros((out_c, f_c, d_bolck_d, d_bolck_h, d_bolck_w))
    filter_dilation[:, :, 0:d_bolck_d:dilation[0], 0:d_bolck_h:dilation[1], 0:
                    d_bolck_w:dilation[2]] = filter

51 52 53 54 55 56
    for d in range(out_d):
        for i in range(out_h):
            for j in range(out_w):
                for g in range(group):
                    input_pad_masked = \
                        input_pad[:, g * f_c:(g + 1) * f_c,
C
chengduoZH 已提交
57 58 59 60 61 62
                        d * stride[0]:d * stride[0] + d_bolck_d,
                        i * stride[1]:i * stride[1] + d_bolck_h,
                        j * stride[2]:j * stride[2] + d_bolck_w]

                    f_sub = filter_dilation[g * sub_out_c:(g + 1) *
                                            sub_out_c, :, :, :, :]
63 64 65
                    for k in range(sub_out_c):
                        out[:, g * sub_out_c + k, d, i, j] = \
                            np.sum(input_pad_masked * f_sub[k, :, :, :, :],
C
chengduoZH 已提交
66
                                   axis=(1, 2, 3, 4))
67 68 69 70

    return out


C
chengduoZH 已提交
71 72
class TestConv3dOp(OpTest):
    def setUp(self):
73
        self.use_cudnn = False
74 75
        self.init_group()
        self.init_op_type()
C
chengduoZH 已提交
76
        self.init_dilation()
77 78
        self.init_test_case()

C
chengduoZH 已提交
79 80 81
        conv3d_param = {
            'stride': self.stride,
            'pad': self.pad,
82 83 84
            'dilations': self.dilations,
            'use_cudnn': self.use_cudnn,
            'data_format': 'AnyLayout'  # TODO(dzhwinter) : should be fix latter
C
chengduoZH 已提交
85
        }
86 87
        input = np.random.random(self.input_size).astype("float32")
        filter = np.random.random(self.filter_size).astype("float32")
C
chengduoZH 已提交
88 89
        output = conv3d_forward_naive(input, filter, self.groups,
                                      conv3d_param).astype("float32")
C
chengduoZH 已提交
90 91 92

        self.inputs = {'Input': input, 'Filter': filter}
        self.attrs = {
93 94
            'strides': self.stride,
            'paddings': self.pad,
C
chengduoZH 已提交
95 96
            'groups': self.groups,
            'dilations': self.dilations
C
chengduoZH 已提交
97 98 99 100
        }
        self.outputs = {'Output': output}

    def test_check_output(self):
101 102 103 104 105
        if self.use_cudnn:
            place = core.CUDAPlace(0)
            self.check_output_with_place(place, atol=1e-5)
        else:
            self.check_output()
C
chengduoZH 已提交
106 107

    def test_check_grad(self):
108 109 110 111 112 113 114 115 116 117
        if self.use_cudnn:
            place = core.CUDAPlace(0)
            self.check_grad_with_place(
                place,
                set(['Input', 'Filter']),
                'Output',
                max_relative_error=0.03)
        else:
            self.check_grad(
                set(['Input', 'Filter']), 'Output', max_relative_error=0.03)
C
chengduoZH 已提交
118

C
chengduoZH 已提交
119
    def test_check_grad_no_filter(self):
120 121 122 123 124 125 126 127 128 129 130 131 132
        if self.use_cudnn:
            place = core.CUDAPlace(0)
            self.check_grad_with_place(
                place, ['Input'],
                'Output',
                max_relative_error=0.03,
                no_grad_set=set(['Filter']))
        else:
            self.check_grad(
                ['Input'],
                'Output',
                max_relative_error=0.03,
                no_grad_set=set(['Filter']))
C
chengduoZH 已提交
133 134

    def test_check_grad_no_input(self):
135 136 137 138 139 140 141 142 143 144 145 146 147
        if self.use_cudnn:
            place = core.CUDAPlace(0)
            self.check_grad_with_place(
                place, ['Filter'],
                'Output',
                max_relative_error=0.03,
                no_grad_set=set(['Input']))
        else:
            self.check_grad(
                ['Filter'],
                'Output',
                max_relative_error=0.03,
                no_grad_set=set(['Input']))
C
chengduoZH 已提交
148

149 150 151
    def init_test_case(self):
        self.pad = [0, 0, 0]
        self.stride = [1, 1, 1]
C
chengduoZH 已提交
152
        self.input_size = [2, 3, 4, 4, 4]  # NCDHW
153 154 155 156
        assert np.mod(self.input_size[1], self.groups) == 0
        f_c = self.input_size[1] / self.groups
        self.filter_size = [6, f_c, 3, 3, 3]

C
chengduoZH 已提交
157 158 159
    def init_dilation(self):
        self.dilations = [1, 1, 1]

160
    def init_group(self):
C
chengduoZH 已提交
161 162
        self.groups = 1

163 164 165
    def init_op_type(self):
        self.op_type = "conv3d"

C
chengduoZH 已提交
166

C
chengduoZH 已提交
167 168 169 170
class TestCase1(TestConv3dOp):
    def init_test_case(self):
        self.pad = [1, 1, 1]
        self.stride = [1, 1, 1]
C
chengduoZH 已提交
171
        self.input_size = [2, 3, 4, 4, 4]  # NCDHW
C
chengduoZH 已提交
172 173 174 175 176
        assert np.mod(self.input_size[1], self.groups) == 0
        f_c = self.input_size[1] / self.groups
        self.filter_size = [6, f_c, 3, 3, 3]


C
chengduoZH 已提交
177 178 179
class TestWithGroup1(TestConv3dOp):
    def init_group(self):
        self.groups = 3
C
chengduoZH 已提交
180 181


C
chengduoZH 已提交
182
class TestWithGroup2(TestCase1):
183
    def init_group(self):
C
chengduoZH 已提交
184 185
        self.groups = 3

186

C
chengduoZH 已提交
187 188 189 190 191 192 193 194 195 196 197
class TestWith1x1(TestConv3dOp):
    def init_test_case(self):
        self.pad = [0, 0, 0]
        self.stride = [1, 1, 1]
        self.input_size = [2, 3, 4, 4, 4]  # NCHW
        assert np.mod(self.input_size[1], self.groups) == 0
        f_c = self.input_size[1] / self.groups
        self.filter_size = [6, f_c, 1, 1, 1]

    def init_dilation(self):
        self.dilations = [1, 1, 1]
C
chengduoZH 已提交
198

C
chengduoZH 已提交
199 200 201
    def init_group(self):
        self.groups = 3

C
chengduoZH 已提交
202 203 204 205 206 207 208 209 210 211 212 213 214 215 216

class TestWithDilation(TestConv3dOp):
    def init_test_case(self):
        self.pad = [0, 0, 0]
        self.stride = [1, 1, 1]
        self.input_size = [2, 3, 6, 6, 6]  # NCDHW
        assert np.mod(self.input_size[1], self.groups) == 0
        f_c = self.input_size[1] / self.groups
        self.filter_size = [6, f_c, 2, 2, 2]

    def init_dilation(self):
        self.dilations = [2, 2, 2]

    def init_group(self):
        self.groups = 3
C
chengduoZH 已提交
217

C
chengduoZH 已提交
218

219
class TestCUDNN(TestConv3dOp):
武毅 已提交
220
    def init_op_type(self):
221 222
        self.use_cudnn = True
        self.op_type = "conv3d"
武毅 已提交
223 224


225
class TestWithGroup1CUDNN(TestWithGroup1):
武毅 已提交
226
    def init_op_type(self):
227 228
        self.use_cudnn = True
        self.op_type = "conv3d"
武毅 已提交
229 230


231
class TestWithGroup2CUDNN(TestWithGroup2):
武毅 已提交
232
    def init_op_type(self):
233 234
        self.use_cudnn = True
        self.op_type = "conv3d"
武毅 已提交
235 236


237
class TestWith1x1CUDNN(TestWith1x1):
武毅 已提交
238
    def init_op_type(self):
239 240
        self.use_cudnn = True
        self.op_type = "conv3d"
武毅 已提交
241 242 243 244


# FIXME(typhoonzero): find a way to determine if
# using cudnn > 6 in python
245
# class TestWithDilationCUDNN(TestWithDilation):
武毅 已提交
246
#     def init_op_type(self):
247
#         self.op_type = "conv3d"
武毅 已提交
248

C
chengduoZH 已提交
249 250
if __name__ == '__main__':
    unittest.main()