bug之torchsummary总结模型出现问题:TypeError: can't multiply sequence by non-int of type 'list'
项目场景：
源码：源码是全卷积神经网络，每一层输出的特征出的宽高一致，只有通道数的变化，即每一层由不同数量和核的卷积构成，最后输出的特征图个数对应多少个关键点。整个网络评判的是，图像中每一个像素位置是某一个关键点的概率，是将图像转换成数值坐标的表示。
但这里谈的并不是这个，是torchsummary在打印模型结构的时候出现了bug
import dsntnn
import torch
from torch import nn, optim
# device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
class FCN(nn.Module):
def __init__(self):
super().__init__()
self.layers = nn.Sequential(
nn.Conv2d(3, 48, kernel_size=7, padding=3),
nn.ReLU(),
nn.Conv2d(48, 48, kernel_size=5, padding=2),
nn.ReLU(),
nn.Conv2d(48, 24, kernel_size=5, padding=2),
nn.ReLU(),
nn.Conv2d(24, 24, kernel_size=3, padding=1),
)
def forward(self, x):
return self.layers(x)
class CoordRegressionNetwork(nn.Module):
def __init__(self, n_locations):
super().__init__()
self.fcn = FCN()
self.hm_conv = nn.Conv2d(24, n_locations, kernel_size=1, bias=False)
def forward(self, images):
# 1. Run the images through our FCN
fcn_out = self.fcn(images)
# 2. Use a 1x1 conv to get one unnormalized heatmap per location
unnormalized_heatmaps = self.hm_conv(fcn_out)
# 3. Normalize the heatmaps
heatmaps = dsntnn.flat_softmax(unnormalized_heatmaps)
# 4. Calculate the coordinates
coords = dsntnn.dsnt(heatmaps)
return coords, heatmaps
def main():
from torchsummary import summary
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
model = CoordRegressionNetwork(n_locations=24)
summary(model.cuda(), (3, 512, 512))
if __name__ == "__main__":
# 测试打印模型
main()
问题描述：
提示：这里描述项目中遇到的问题：TypeError: can’t multiply sequence by non-int of type 'list’出现再torchsummary。
尺寸大小固定的网络：
在anaconda3+python3.7 +pytorch1.5.1+torchsummary环境下运行正常。
在新环境运行报错，新环境python=3.8.3
File "D:/TanHaiyan/Python_code/KeyPointsDetection_with_Torch/creat_FCN.py", line 71, in main
summary(model.cuda(), (3, 512, 512))
File "D:\TanHaiyan\Program Files\Anaconda3\lib\site-packages\torchsummary-1.5.1-py3.8.egg\torchsummary\torchsummary.py", line 10, in summary
File "D:\TanHaiyan\Program Files\Anaconda3\lib\site-packages\torchsummary-1.5.1-py3.8.egg\torchsummary\torchsummary.py", line 94, in summary_string
File "<__array_function__ internals>", line 5, in prod
File "D:\TanHaiyan\Program Files\Anaconda3\lib\site-packages\numpy\core\fromnumeric.py", line 2961, in prod
return _wrapreduction(a, np.multiply, 'prod', axis, dtype, out,
File "D:\TanHaiyan\Program Files\Anaconda3\lib\site-packages\numpy\core\fromnumeric.py", line 90, in _wrapreduction
return ufunc.reduce(obj, axis, dtype, out, **passkwargs)
TypeError: can't multiply sequence by non-int of type 'list'
原因分析：
1.环境不匹配。
2.模型输出两个量，两者之间传递信息可能不是nn.Module的子类。
解决方案：
方法一：将环境换回python3.7
方法二：采用新的方法读取模型，采用下列代码段替代main函数的函数体：
# Initialize model
# model = TheModelClass()
model = CoordRegressionNetwork(n_locations=24)
# Initialize optimizer
optimizer = torch.optim.SGD(model.parameters(), lr=1e-4, momentum=0.9)
print("Model's state_dict:")
# Print model's state_dict
for param_tensor in model.state_dict():
print(param_tensor, "\t", model.state_dict()[param_tensor].size())
print("optimizer's state_dict:")
# Print optimizer's state_dict
for var_name in optimizer.state_dict():
print(var_name, "\t", optimizer.state_dict()[var_name])
方法三：针对分析原因2，去掉一个输出。模型能正常总结summary
修改模型最后输出：
输出：
----------------------------------------------------------------
Layer (type) Output Shape Param #
================================================================
Conv2d-1 [-1, 48, 512, 512] 7,104
ReLU-2 [-1, 48, 512, 512] 0
Conv2d-3 [-1, 48, 512, 512] 57,648
ReLU-4 [-1, 48, 512, 512] 0
Conv2d-5 [-1, 24, 512, 512] 28,824
ReLU-6 [-1, 24, 512, 512] 0
Conv2d-7 [-1, 24, 512, 512] 5,208
FCN-8 [-1, 24, 512, 512] 0
Conv2d-9 [-1, 24, 512, 512] 576
CoordRegressionNetwork-10 [-1, 24, 512, 512] 0
================================================================
Total params: 99,360
Trainable params: 99,360
Non-trainable params: 0
----------------------------------------------------------------
Input size (MB): 3.00
Forward/backward pass size (MB): 672.00
Params size (MB): 0.38
Estimated Total Size (MB): 675.38
----------------------------------------------------------------
Process finished with exit code 0
