目录
一、resnet部分加速理解
-
网络整体流向
在ResNet类中的forward( )函数规定了网络数据的流向：
（1）数据进入网络后先经过输入部分（conv1, bn1, relu, maxpool）；
（2）然后进入中间卷积部分（layer1, layer2, layer3, layer4，这里的layer对应我们之前所说的stage）；
（3）最后数据经过一个平均池化和全连接层（avgpool, fc）输出得到结果；
具体来说，resnet50和其他res系列网络的差异主要在于layer1~layer4，其他的部件都是相似的。 -
网络输入部分详解：
所有的ResNet网络输入部分是一个size=7x7, stride=2的大卷积核，以及一个size=3x3, stride=2的最大池化组成，通过这一步，一个224x224的输入图像就会变56x56大小的特征图，极大减少了存储所需大小。 -
网络中间卷积部分
中间卷积部分主要是下图中的蓝框部分，通过3*3卷积的堆叠来实现信息的提取。红框中的[2, 2, 2, 2]和[3, 4, 6, 3]等则代表了bolck的重复堆叠次数。[3, 4, 6, 3]代表resnet-34，同理，[3, 4, 23, 3]代表resnet-101.
resnet残差部分可参考：PyTorch—torchvision.models导入预训练模型—残差网络代码讲解
此时，以resnet50为例，查看神经网络的形状和参数大小，主代码如下：
from torchsummary import summary
import torchvision.models as models
model = models.resnet50()
summary(model, input_size=(3,256,256), batch_size=-1, device='cuda')打印出来的结果如下所示：
C:\ProgramData\Anaconda3\python.exe F:/lingjun2019/Body_Recog/study_test/print_netParam.py
----------------------------------------------------------------
Layer (type) Output Shape Param #
================================================================
Conv2d-1 [-1, 64, 112, 112] 9,408
BatchNorm2d-2 [-1, 64, 112, 112] 128
ReLU-3 [-1, 64, 112, 112] 0
MaxPool2d-4 [-1, 64, 56, 56] 0
Conv2d-5 [-1, 64, 56, 56] 4,096
BatchNorm2d-6 [-1, 64, 56, 56] 128
ReLU-7 [-1, 64, 56, 56] 0
Conv2d-8 [-1, 64, 56, 56] 36,864
BatchNorm2d-9 [-1, 64, 56, 56] 128
ReLU-10 [-1, 64, 56, 56] 0
Conv2d-11 [-1, 256, 56, 56] 16,384
BatchNorm2d-12 [-1, 256, 56, 56] 512
Conv2d-13 [-1, 256, 56, 56] 16,384
BatchNorm2d-14 [-1, 256, 56, 56] 512
ReLU-15 [-1, 256, 56, 56] 0
Bottleneck-16 [-1, 256, 56, 56] 0
Conv2d-17 [-1, 64, 56, 56] 16,384
BatchNorm2d-18 [-1, 64, 56, 56] 128
ReLU-19 [-1, 64, 56, 56] 0
Conv2d-20 [-1, 64, 56, 56] 36,864
BatchNorm2d-21 [-1, 64, 56, 56] 128
ReLU-22 [-1, 64, 56, 56] 0
Conv2d-23 [-1, 256, 56, 56] 16,384
BatchNorm2d-24 [-1, 256, 56, 56] 512
ReLU-25 [-1, 256, 56, 56] 0
Bottleneck-26 [-1, 256, 56, 56] 0
Conv2d-27 [-1, 64, 56, 56] 16,384
BatchNorm2d-28 [-1, 64, 56, 56] 128
ReLU-29 [-1, 64, 56, 56] 0
Conv2d-30 [-1, 64, 56, 56] 36,864
BatchNorm2d-31 [-1, 64, 56, 56] 128
ReLU-32 [-1, 64, 56, 56] 0
Conv2d-33 [-1, 256, 56, 56] 16,384
BatchNorm2d-34 [-1, 256, 56, 56] 512
ReLU-35 [-1, 256, 56, 56] 0
Bottleneck-36 [-1, 256, 56, 56] 0
Conv2d-37 [-1, 128, 56, 56] 32,768
BatchNorm2d-38 [-1, 128, 56, 56] 256
ReLU-39 [-1, 128, 56, 56] 0
Conv2d-40 [-1, 128, 28, 28] 147,456
BatchNorm2d-41 [-1, 128, 28, 28] 256
ReLU-42 [-1, 128, 28, 28] 0
Conv2d-43 [-1, 512, 28, 28] 65,536
BatchNorm2d-44 [-1, 512, 28, 28] 1,024
Conv2d-45 [-1, 512, 28, 28] 131,072
BatchNorm2d-46 [-1, 512, 28, 28] 1,024
ReLU-47 [-1, 512, 28, 28] 0
Bottleneck-48 [-1, 512, 28, 28] 0
Conv2d-49 [-1, 128, 28, 28] 65,536
BatchNorm2d-50 [-1, 128, 28, 28] 256
ReLU-51 [-1, 128, 28, 28] 0
Conv2d-52 [-1, 128, 28, 28] 147,456
BatchNorm2d-53 [-1, 128, 28, 28] 256
ReLU-54 [-1, 128, 28, 28] 0
Conv2d-55 [-1, 512, 28, 28] 65,536
BatchNorm2d-56 [-1, 512, 28, 28] 1,024
ReLU-57 [-1, 512, 28, 28] 0
Bottleneck-58 [-1, 512, 28, 28] 0
Conv2d-59 [-1, 128, 28, 28] 65,536
BatchNorm2d-60 [-1, 128, 28, 28] 256
ReLU-61 [-1, 128, 28, 28] 0
Conv2d-62 [-1, 128, 28, 28] 147,456
BatchNorm2d-63 [-1, 128, 28, 28] 256
ReLU-64 [-1, 128, 28, 28] 0
Conv2d-65 [-1, 512, 28, 28] 65,536
BatchNorm2d-66 [-1, 512, 28, 28] 1,024
ReLU-67 [-1, 512, 28, 28] 0
Bottleneck-68 [-1, 512, 28, 28] 0
Conv2d-69 [-1, 128, 28, 28] 65,536
BatchNorm2d-70 [-1, 128, 28, 28] 256
ReLU-71 [-1, 128, 28, 28] 0
Conv2d-72 [-1, 128, 28, 28] 147,456
BatchNorm2d-73 [-1, 128, 28, 28] 256
ReLU-74 [-1, 128, 28, 28] 0
Conv2d-75 [-1, 512, 28, 28] 65,536
BatchNorm2d-76 [-1, 512, 28, 28] 1,024
ReLU-77 [-1, 512, 28, 28] 0
Bottleneck-78 [-1, 512, 28, 28] 0
Conv2d-79 [-1, 256, 28, 28] 131,072
BatchNorm2d-80 [-1, 256, 28, 28] 512
ReLU-81 [-1, 256, 28, 28] 0
Conv2d-82 [-1, 256, 14, 14] 589,824
BatchNorm2d-83 [-1, 256, 14, 14] 512
ReLU-84 [-1, 256, 14, 14] 0
Conv2d-85 [-1, 1024, 14, 14] 262,144
BatchNorm2d-86 [-1, 1024, 14, 14] 2,048
Conv2d-87 [-1, 1024, 14, 14] 524,288
BatchNorm2d-88 [-1, 1024, 14, 14] 2,048
ReLU-89 [-1, 1024, 14, 14] 0
Bottleneck-90 [-1, 1024, 14, 14] 0
Conv2d-91 [-1, 256, 14, 14] 262,144
BatchNorm2d-92 [-1, 256, 14, 14] 512
ReLU-93 [-1, 256, 14, 14] 0
Conv2d-94 [-1, 256, 14, 14] 589,824
BatchNorm2d-95 [-1, 256, 14, 14] 512
ReLU-96 [-1, 256, 14, 14] 0
Conv2d-97 [-1, 1024, 14, 14] 262,144
BatchNorm2d-98 [-1, 1024, 14, 14] 2,048
ReLU-99 [-1, 1024, 14, 14] 0
Bottleneck-100 [-1, 1024, 14, 14] 0
Conv2d-101 [-1, 256, 14, 14] 262,144
BatchNorm2d-102 [-1, 256, 14, 14] 512
ReLU-103 [-1, 256, 14, 14] 0
Conv2d-104 [-1, 256, 14, 14] 589,824
BatchNorm2d-105 [-1, 256, 14, 14] 512
ReLU-106 [-1, 256, 14, 14] 0
Conv2d-107 [-1, 1024, 14, 14] 262,144
BatchNorm2d-108 [-1, 1024, 14, 14] 2,048
ReLU-109 [-1, 1024, 14, 14] 0
Bottleneck-110 [-1, 1024, 14, 14] 0
Conv2d-111 [-1, 256, 14, 14] 262,144
BatchNorm2d-112 [-1, 256, 14, 14] 512
ReLU-113 [-1, 256, 14, 14] 0
Conv2d-114 [-1, 256, 14, 14] 589,824
BatchNorm2d-115 [-1, 256, 14, 14] 512
ReLU-116 [-1, 256, 14, 14] 0
Conv2d-117 [-1, 1024, 14, 14] 262,144
BatchNorm2d-118 [-1, 1024, 14, 14] 2,048
ReLU-119 [-1, 1024, 14, 14] 0
Bottleneck-120 [-1, 1024, 14, 14] 0
Conv2d-121 [-1, 256, 14, 14] 262,144
BatchNorm2d-122 [-1, 256, 14, 14] 512
ReLU-123 [-1, 256, 14, 14] 0
Conv2d-124 [-1, 256, 14, 14] 589,824
BatchNorm2d-125 [-1, 256, 14, 14] 512
ReLU-126 [-1, 256, 14, 14] 0
Conv2d-127 [-1, 1024, 14, 14] 262,144
BatchNorm2d-128 [-1, 1024, 14, 14] 2,048
ReLU-129 [-1, 1024, 14, 14] 0
Bottleneck-130 [-1, 1024, 14, 14] 0
Conv2d-131 [-1, 256, 14, 14] 262,144
BatchNorm2d-132 [-1, 256, 14, 14] 512
ReLU-133 [-1, 256, 14, 14] 0
Conv2d-134 [-1, 256, 14, 14] 589,824
BatchNorm2d-135 [-1, 256, 14, 14] 512
ReLU-136 [-1, 256, 14, 14] 0
Conv2d-137 [-1, 1024, 14, 14] 262,144
BatchNorm2d-138 [-1, 1024, 14, 14] 2,048
ReLU-139 [-1, 1024, 14, 14] 0
Bottleneck-140 [-1, 1024, 14, 14] 0
Conv2d-141 [-1, 256, 14, 14] 262,144
BatchNorm2d-142 [-1, 256, 14, 14] 512
ReLU-143 [-1, 256, 14, 14] 0
Conv2d-144 [-1, 256, 14, 14] 589,824
BatchNorm2d-145 [-1, 256, 14, 14] 512
ReLU-146 [-1, 256, 14, 14] 0
Conv2d-147 [-1, 1024, 14, 14] 262,144
BatchNorm2d-148 [-1, 1024, 14, 14] 2,048
ReLU-149 [-1, 1024, 14, 14] 0
Bottleneck-150 [-1, 1024, 14, 14] 0
Conv2d-151 [-1, 256, 14, 14] 262,144
BatchNorm2d-152 [-1, 256, 14, 14] 512
ReLU-153 [-1, 256, 14, 14] 0
Conv2d-154 [-1, 256, 14, 14] 589,824
BatchNorm2d-155 [-1, 256, 14, 14] 512
ReLU-156 [-1, 256, 14, 14] 0
Conv2d-157 [-1, 1024, 14, 14] 262,144
BatchNorm2d-158 [-1, 1024, 14, 14] 2,048
ReLU-159 [-1, 1024, 14, 14] 0
Bottleneck-160 [-1, 1024, 14, 14] 0
Conv2d-161 [-1, 256, 14, 14] 262,144
BatchNorm2d-162 [-1, 256, 14, 14] 512
ReLU-163 [-1, 256, 14, 14] 0
Conv2d-164 [-1, 256, 14, 14] 589,824
BatchNorm2d-165 [-1, 256, 14, 14] 512
ReLU-166 [-1, 256, 14, 14] 0
Conv2d-167 [-1, 1024, 14, 14] 262,144
BatchNorm2d-168 [-1, 1024, 14, 14] 2,048
ReLU-169 [-1, 1024, 14, 14] 0
Bottleneck-170 [-1, 1024, 14, 14] 0
Conv2d-171 [-1, 256, 14, 14] 262,144
BatchNorm2d-172 [-1, 256, 14, 14] 512
ReLU-173 [-1, 256, 14, 14] 0
Conv2d-174 [-1, 256, 14, 14] 589,824
BatchNorm2d-175 [-1, 256, 14, 14] 512
ReLU-176 [-1, 256, 14, 14] 0
Conv2d-177 [-1, 1024, 14, 14] 262,144
BatchNorm2d-178 [-1, 1024, 14, 14] 2,048
ReLU-179 [-1, 1024, 14, 14] 0
Bottleneck-180 [-1, 1024, 14, 14] 0
Conv2d-181 [-1, 256, 14, 14] 262,144
BatchNorm2d-182 [-1, 256, 14, 14] 512
ReLU-183 [-1, 256, 14, 14] 0
Conv2d-184 [-1, 256, 14, 14] 589,824
BatchNorm2d-185 [-1, 256, 14, 14] 512
ReLU-186 [-1, 256, 14, 14] 0
Conv2d-187 [-1, 1024, 14, 14] 262,144
BatchNorm2d-188 [-1, 1024, 14, 14] 2,048
ReLU-189 [-1, 1024, 14, 14] 0
Bottleneck-190 [-1, 1024, 14, 14] 0
Conv2d-191 [-1, 256, 14, 14] 262,144
BatchNorm2d-192 [-1, 256, 14, 14] 512
ReLU-193 [-1, 256, 14, 14] 0
Conv2d-194 [-1, 256, 14, 14] 589,824
BatchNorm2d-195 [-1, 256, 14, 14] 512
ReLU-196 [-1, 256, 14, 14] 0
Conv2d-197 [-1, 1024, 14, 14] 262,144
BatchNorm2d-198 [-1, 1024, 14, 14] 2,048
ReLU-199 [-1, 1024, 14, 14] 0
Bottleneck-200 [-1, 1024, 14, 14] 0
Conv2d-201 [-1, 256, 14, 14] 262,144
BatchNorm2d-202 [-1, 256, 14, 14] 512
ReLU-203 [-1, 256, 14, 14] 0
Conv2d-204 [-1, 256, 14, 14] 589,824
BatchNorm2d-205 [-1, 256, 14, 14] 512
ReLU-206 [-1, 256, 14, 14] 0
Conv2d-207 [-1, 1024, 14, 14] 262,144
BatchNorm2d-208 [-1, 1024, 14, 14] 2,048
ReLU-209 [-1, 1024, 14, 14] 0
Bottleneck-210 [-1, 1024, 14, 14] 0
Conv2d-211 [-1, 256, 14, 14] 262,144
BatchNorm2d-212 [-1, 256, 14, 14] 512
ReLU-213 [-1, 256, 14, 14] 0
Conv2d-214 [-1, 256, 14, 14] 589,824
BatchNorm2d-215 [-1, 256, 14, 14] 512
ReLU-216 [-1, 256, 14, 14] 0
Conv2d-217 [-1, 1024, 14, 14] 262,144
BatchNorm2d-218 [-1, 1024, 14, 14] 2,048
ReLU-219 [-1, 1024, 14, 14] 0
Bottleneck-220 [-1, 1024, 14, 14] 0
Conv2d-221 [-1, 256, 14, 14] 262,144
BatchNorm2d-222 [-1, 256, 14, 14] 512
ReLU-223 [-1, 256, 14, 14] 0
Conv2d-224 [-1, 256, 14, 14] 589,824
BatchNorm2d-225 [-1, 256, 14, 14] 512
ReLU-226 [-1, 256, 14, 14] 0
Conv2d-227 [-1, 1024, 14, 14] 262,144
BatchNorm2d-228 [-1, 1024, 14, 14] 2,048
ReLU-229 [-1, 1024, 14, 14] 0
Bottleneck-230 [-1, 1024, 14, 14] 0
Conv2d-231 [-1, 256, 14, 14] 262,144
BatchNorm2d-232 [-1, 256, 14, 14] 512
ReLU-233 [-1, 256, 14, 14] 0
Conv2d-234 [-1, 256, 14, 14] 589,824
BatchNorm2d-235 [-1, 256, 14, 14] 512
ReLU-236 [-1, 256, 14, 14] 0
Conv2d-237 [-1, 1024, 14, 14] 262,144
BatchNorm2d-238 [-1, 1024, 14, 14] 2,048
ReLU-239 [-1, 1024, 14, 14] 0
Bottleneck-240 [-1, 1024, 14, 14] 0
Conv2d-241 [-1, 256, 14, 14] 262,144
BatchNorm2d-242 [-1, 256, 14, 14] 512
ReLU-243 [-1, 256, 14, 14] 0
Conv2d-244 [-1, 256, 14, 14] 589,824
BatchNorm2d-245 [-1, 256, 14, 14] 512
ReLU-246 [-1, 256, 14, 14] 0
Conv2d-247 [-1, 1024, 14, 14] 262,144
BatchNorm2d-248 [-1, 1024, 14, 14] 2,048
ReLU-249 [-1, 1024, 14, 14] 0
Bottleneck-250 [-1, 1024, 14, 14] 0
Conv2d-251 [-1, 256, 14, 14] 262,144
BatchNorm2d-252 [-1, 256, 14, 14] 512
ReLU-253 [-1, 256, 14, 14] 0
Conv2d-254 [-1, 256, 14, 14] 589,824
BatchNorm2d-255 [-1, 256, 14, 14] 512
ReLU-256 [-1, 256, 14, 14] 0
Conv2d-257 [-1, 1024, 14, 14] 262,144
BatchNorm2d-258 [-1, 1024, 14, 14] 2,048
ReLU-259 [-1, 1024, 14, 14] 0
Bottleneck-260 [-1, 1024, 14, 14] 0
Conv2d-261 [-1, 256, 14, 14] 262,144
BatchNorm2d-262 [-1, 256, 14, 14] 512
ReLU-263 [-1, 256, 14, 14] 0
Conv2d-264 [-1, 256, 14, 14] 589,824
BatchNorm2d-265 [-1, 256, 14, 14] 512
ReLU-266 [-1, 256, 14, 14] 0
Conv2d-267 [-1, 1024, 14, 14] 262,144
BatchNorm2d-268 [-1, 1024, 14, 14] 2,048
ReLU-269 [-1, 1024, 14, 14] 0
Bottleneck-270 [-1, 1024, 14, 14] 0
Conv2d-271 [-1, 256, 14, 14] 262,144
BatchNorm2d-272 [-1, 256, 14, 14] 512
ReLU-273 [-1, 256, 14, 14] 0
Conv2d-274 [-1, 256, 14, 14] 589,824
BatchNorm2d-275 [-1, 256, 14, 14] 512
ReLU-276 [-1, 256, 14, 14] 0
Conv2d-277 [-1, 1024, 14, 14] 262,144
BatchNorm2d-278 [-1, 1024, 14, 14] 2,048
ReLU-279 [-1, 1024, 14, 14] 0
Bottleneck-280 [-1, 1024, 14, 14] 0
Conv2d-281 [-1, 256, 14, 14] 262,144
BatchNorm2d-282 [-1, 256, 14, 14] 512
ReLU-283 [-1, 256, 14, 14] 0
Conv2d-284 [-1, 256, 14, 14] 589,824
BatchNorm2d-285 [-1, 256, 14, 14] 512
ReLU-286 [-1, 256, 14, 14] 0
Conv2d-287 [-1, 1024, 14, 14] 262,144
BatchNorm2d-288 [-1, 1024, 14, 14] 2,048
ReLU-289 [-1, 1024, 14, 14] 0
Bottleneck-290 [-1, 1024, 14, 14] 0
Conv2d-291 [-1, 256, 14, 14] 262,144
BatchNorm2d-292 [-1, 256, 14, 14] 512
ReLU-293 [-1, 256, 14, 14] 0
Conv2d-294 [-1, 256, 14, 14] 589,824
BatchNorm2d-295 [-1, 256, 14, 14] 512
ReLU-296 [-1, 256, 14, 14] 0
Conv2d-297 [-1, 1024, 14, 14] 262,144
BatchNorm2d-298 [-1, 1024, 14, 14] 2,048
ReLU-299 [-1, 1024, 14, 14] 0
Bottleneck-300 [-1, 1024, 14, 14] 0
Conv2d-301 [-1, 256, 14, 14] 262,144
BatchNorm2d-302 [-1, 256, 14, 14] 512
ReLU-303 [-1, 256, 14, 14] 0
Conv2d-304 [-1, 256, 14, 14] 589,824
BatchNorm2d-305 [-1, 256, 14, 14] 512
ReLU-306 [-1, 256, 14, 14] 0
Conv2d-307 [-1, 1024, 14, 14] 262,144
BatchNorm2d-308 [-1, 1024, 14, 14] 2,048
ReLU-309 [-1, 1024, 14, 14] 0
Bottleneck-310 [-1, 1024, 14, 14] 0
Conv2d-311 [-1, 512, 14, 14] 524,288
BatchNorm2d-312 [-1, 512, 14, 14] 1,024
ReLU-313 [-1, 512, 14, 14] 0
Conv2d-314 [-1, 512, 7, 7] 2,359,296
BatchNorm2d-315 [-1, 512, 7, 7] 1,024
ReLU-316 [-1, 512, 7, 7] 0
Conv2d-317 [-1, 2048, 7, 7] 1,048,576
BatchNorm2d-318 [-1, 2048, 7, 7] 4,096
Conv2d-319 [-1, 2048, 7, 7] 2,097,152
BatchNorm2d-320 [-1, 2048, 7, 7] 4,096
ReLU-321 [-1, 2048, 7, 7] 0
Bottleneck-322 [-1, 2048, 7, 7] 0
Conv2d-323 [-1, 512, 7, 7] 1,048,576
BatchNorm2d-324 [-1, 512, 7, 7] 1,024
ReLU-325 [-1, 512, 7, 7] 0
Conv2d-326 [-1, 512, 7, 7] 2,359,296
BatchNorm2d-327 [-1, 512, 7, 7] 1,024
ReLU-328 [-1, 512, 7, 7] 0
Conv2d-329 [-1, 2048, 7, 7] 1,048,576
BatchNorm2d-330 [-1, 2048, 7, 7] 4,096
ReLU-331 [-1, 2048, 7, 7] 0
Bottleneck-332 [-1, 2048, 7, 7] 0
Conv2d-333 [-1, 512, 7, 7] 1,048,576
BatchNorm2d-334 [-1, 512, 7, 7] 1,024
ReLU-335 [-1, 512, 7, 7] 0
Conv2d-336 [-1, 512, 7, 7] 2,359,296
BatchNorm2d-337 [-1, 512, 7, 7] 1,024
ReLU-338 [-1, 512, 7, 7] 0
Conv2d-339 [-1, 2048, 7, 7] 1,048,576
BatchNorm2d-340 [-1, 2048, 7, 7] 4,096
ReLU-341 [-1, 2048, 7, 7] 0
Bottleneck-342 [-1, 2048, 7, 7] 0
AdaptiveAvgPool2d-343 [-1, 2048, 1, 1] 0
Linear-344 [-1, 1000] 2,049,000
================================================================
Total params: 44,549,160
Trainable params: 44,549,160
Non-trainable params: 0
----------------------------------------------------------------
Input size (MB): 0.57
Forward/backward pass size (MB): 429.73
Params size (MB): 169.94
Estimated Total Size (MB): 600.25
----------------------------------------------------------------
Process finished with exit code 0
如何将这个输出内容，保存到本地的TXT文件？
当你使用torch.summary函数来输出模型的形状和参数大小时，你可以通过将标准输出（stdout）重定向到本地TXT文件来保存输出内容。Python提供了一个上下文管理器with open()，可以帮助我们实现这一功能，将输出写入文件而不在控制台上显示。
以下是实现方法：
import sys
from contextlib import contextmanager
@contextmanager
def stdout_redirected(to=None):
"""
上下文管理器，用于临时将stdout重定向到文件或控制台。
使用方法：`with stdout_redirected(to='output.txt'):`
"""
if to is None:
yield
else:
sys.stdout.flush()
original_stdout = sys.stdout
with open(to, 'w') as file:
sys.stdout = file
try:
yield file
finally:
sys.stdout = original_stdout
# 假设你已经创建了PyTorch模型，并将其命名为`model`
# 下面演示如何使用`torch.summary`函数并将输出保存到文件
from torchsummary import summary
output_file = "model_summary.txt"
with stdout_redirected(to=output_file):
summary(model, input_size=(input_channels, input_height, input_width))
print(f"模型摘要已保存到{output_file}文件中。")
在这段代码中，我们定义了一个上下文管理器stdout_redirected，它可以临时将标准输出(sys.stdout)重定向到一个文件（如果提供了to参数），或者将其恢复到控制台（如果to是None）。然后，我们使用这个上下文管理器来捕获torch.summary的输出，并将其保存到指定的文件model_summary.txt中。
确保将model替换为你实际的PyTorch模型，并根据你的模型输入形状为input_channels、input_height和input_width设置适当的值。运行代码后，你应该会在model_summary.txt文件中看到模型摘要信息。
为了便于修改，我们换一种方式，打印下model看看，如下这样：
from torchsummary import summary
import torchvision.models as models
model = models.resnet50()
print(model)
# summary(model, input_size=(3, 256, 256), batch_size=-1, device='cuda')打印的结果是下面这样的，就是model的结构：
C:\ProgramData\Anaconda3\python.exe F:/lingjun2019/Body_Recog/study_test/print_netParam.py
ResNet(
(conv1): Conv2d(3, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3), bias=False)
(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(maxpool): MaxPool2d(kernel_size=3, stride=2, padding=1, dilation=1, ceil_mode=False)
(layer1): Sequential(
(0): Bottleneck(
(conv1): Conv2d(64, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(downsample): Sequential(
(0): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): Bottleneck(
(conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(2): Bottleneck(
(conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(layer2): Sequential(
(0): Bottleneck(
(conv1): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(downsample): Sequential(
(0): Conv2d(256, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)
(1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): Bottleneck(
(conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(2): Bottleneck(
(conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(3): Bottleneck(
(conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(layer3): Sequential(
(0): Bottleneck(
(conv1): Conv2d(512, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(downsample): Sequential(
(0): Conv2d(512, 1024, kernel_size=(1, 1), stride=(2, 2), bias=False)
(1): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(2): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(3): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(4): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(5): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(layer4): Sequential(
(0): Bottleneck(
(conv1): Conv2d(1024, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(downsample): Sequential(
(0): Conv2d(1024, 2048, kernel_size=(1, 1), stride=(2, 2), bias=False)
(1): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): Bottleneck(
(conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(2): Bottleneck(
(conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(avgpool): AdaptiveAvgPool2d(output_size=(1, 1))
(fc): Linear(in_features=2048, out_features=1000, bias=True)
)
Process finished with exit code 0
最后全连接层的out_features的数量，就是输出的类别数量，这里默认是1000。假若我们自己的类别只有23个，那这里就需要做个修改，修改方式如下：
from torchsummary import summary
import torchvision.models as models
import torch
model = models.resnet50()
num_frts = model.fc.in_features
model.fc = torch.nn.Linear(num_frts, 23)
print(model)
# summary(model, input_size=(3, 256, 256), batch_size=-1, device='cuda')前面不变，最后的输出的结果变成了下面这样，是我们所希望的：
···
(2): Bottleneck(
(conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(avgpool): AdaptiveAvgPool2d(output_size=(1, 1))
(fc): Linear(in_features=2048, out_features=23, bias=True)
)更多关于分类的详解内容，欢迎订阅我的gitchat： https://gitbook.cn/new/gitchat/activity/603f4254494aab1ce4f231cdGitChat 是一款基于微信平台的知识分享产品。通过这款产品我们希望改变IT知识的学习方式。
https://gitbook.cn/new/gitchat/activity/603f4254494aab1ce4f231cd
二、全局pool辅助理解
ROI pooling是目标检测中常用的一步操作，尤其是在faster rcnn两段式目标检测中。由于是在经过anchor框选的待进一步分类的目标的尺寸大小是不一致的（汽车远近都是汽车，灯塔高高瘦瘦，橘猫又肥又圆），但是简单分类网络的全连接层要求输入尺寸是一致的，这种矛盾就需要一个组件，在中间起到调和的作用。在全连接前，无论什么尺寸，都给统一到一样的大小。
这点我的理解有些类似于一个压榨机。压榨机的出口筛子是固定输出7*7，共49条，并且这个筛子的长宽是可变的，但是，7*7=49个网格不会变。
那长宽尺寸一遍，每一个小格子的尺寸也会发生改变，但始终保持49条输出的这个特性不能变。把它理解成一个中间组件就可以的，如果你有更深入的见解，欢迎评论。
下面是改变input_size，查看网络形状的改变。无论输入图像大小如何变化，到AdaptiveAvgPool2d这一层都需要转化为统一大小。这里就是形如其名的Adaptive，自适应的改变pool形式大小。无论是224的7*7 pool为1*1，还是448的14*14 pool成1*1。（PS，AdaptiveAvgPool2d的详细介绍，帮助理解。：Pytorch 里 nn.AdaptiveAvgPool2d(output_size) 原理是什么?https://www.zhihu.com/question/282046628
除此之外，还引申以下Pool形式进行理解：RoIPooling、RoIAlign笔记
三、总结
torch.summary 这个工具，可以直接打印出神经网络的形状和参数大小，不需要一个节点一个节点的进行打印分析。这样对于我们自己构建出来的模型，查看结构是否和我们预期的一致，以及帮助理解开源的模型结果有非常大的帮助。
torchsummary 的 summary 函数不直接支持多输入模型，建议使用 torchinfo 库来查看模型摘要，它通常对复杂模型有更好的兼容性：
from torchinfo import summary
# 使用 torchinfo
summary(model,
input_size=[(batch, 96, 7), (batch, 96, 4), (batch, 72, 7), (batch, 72, 4)],
device=device)torchinfo 的更多信息，可参考：https://gitcode.com/gh_mirrors/to/torchinfo/overview?utm_source=replace_article_gitcode&index=bottom&type=card&&isLogin=1
最后，如果您觉得本篇文章对你有帮助，欢迎点赞，让更多人看到，这是对我继续写下去的鼓励。如果能再点击下方的红包打赏，给博主来一杯咖啡，那就太好。
参考内容：
.1使用torchsummary打印出神经网络的形状和参数大小https://blog.youkuaiyun.com/sinat_41026716/article/details/105726239
基于PyTorch的卷积神经网络经典BackBone(骨干网络)复现https://mp.weixin.qq.com/s/-t77D71Uu3OLDyUa_a4Yrg
扫一扫
926
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
