pytorch如何获得模型的计算量和参数量_Python

方法1 自带

pytorch自带方法，计算模型参数总量

				?

									total = sum([param.nelement() for param in model.parameters()])

									print("Number of parameter: %.2fM" % (total/1e6))

或者

				?

									total = sum(p.numel() for p in model.parameters())

									print("Total params: %.2fM" % (total/1e6))

方法2 编写代码

计算模型参数总量和模型计算量

				?

									def count_params(model, input_size=224):

									    # param_sum = 0

									    with open('models.txt', 'w') as fm:

									        fm.write(str(model))

									    # 计算模型的计算量

									    calc_flops(model, input_size)

									    # 计算模型的参数总量

									    model_parameters = filter(lambda p: p.requires_grad, model.parameters())

									    params = sum([np.prod(p.size()) for p in model_parameters])

									    print('The network has {} params.'.format(params)) 

									# 计算模型的计算量

									def calc_flops(model, input_size):

									    def conv_hook(self, input, output):

									        batch_size, input_channels, input_height, input_width = input[0].size()

									        output_channels, output_height, output_width = output[0].size()

									        kernel_ops = self.kernel_size[0] * self.kernel_size[1] * (self.in_channels / self.groups) * (

									            2 if multiply_adds else 1)

									        bias_ops = 1 if self.bias is not None else 0

									        params = output_channels * (kernel_ops + bias_ops)

									        flops = batch_size * params * output_height * output_width

									        list_conv.append(flops)

									    def linear_hook(self, input, output):

									        batch_size = input[0].size(0) if input[0].dim() == 2 else 1

									        weight_ops = self.weight.nelement() * (2 if multiply_adds else 1)

									        bias_ops = self.bias.nelement()

									        flops = batch_size * (weight_ops + bias_ops)

									        list_linear.append(flops)

									    def bn_hook(self, input, output):

									        list_bn.append(input[0].nelement())

									    def relu_hook(self, input, output):

									        list_relu.append(input[0].nelement())

									    def pooling_hook(self, input, output):

									        batch_size, input_channels, input_height, input_width = input[0].size()

									        output_channels, output_height, output_width = output[0].size()

									        kernel_ops = self.kernel_size * self.kernel_size

									        bias_ops = 0

									        params = output_channels * (kernel_ops + bias_ops)

									        flops = batch_size * params * output_height * output_width

									        list_pooling.append(flops)

									    def foo(net):

									        childrens = list(net.children())

									        if not childrens:

									            if isinstance(net, torch.nn.Conv2d):

									                net.register_forward_hook(conv_hook)

									            if isinstance(net, torch.nn.Linear):

									                net.register_forward_hook(linear_hook)

									            if isinstance(net, torch.nn.BatchNorm2d):

									                net.register_forward_hook(bn_hook)

									            if isinstance(net, torch.nn.ReLU):

									                net.register_forward_hook(relu_hook)

									            if isinstance(net, torch.nn.MaxPool2d) or isinstance(net, torch.nn.AvgPool2d):

									                net.register_forward_hook(pooling_hook)

									            return

									        for c in childrens:

									            foo(c)

									    multiply_adds = False

									    list_conv, list_bn, list_relu, list_linear, list_pooling = [], [], [], [], []

									    foo(model)

									    if '0.4.' in torch.__version__:

									        if assets.USE_GPU:

									            input = torch.cuda.FloatTensor(torch.rand(2, 3, input_size, input_size).cuda())

									        else:

									            input = torch.FloatTensor(torch.rand(2, 3, input_size, input_size))

									    else:

									        input = Variable(torch.rand(2, 3, input_size, input_size), requires_grad=True)

									    _ = model(input)

									    total_flops = (sum(list_conv) + sum(list_linear) + sum(list_bn) + sum(list_relu) + sum(list_pooling))

									    print('  + Number of FLOPs: %.2fM' % (total_flops / 1e6 / 2))

方法3 thop

需要安装thop

				?

									pip install thop

调用方法：计算模型参数总量和模型计算量，而且会打印每一层网络的具体信息

				?

									from thop import profile 

									input = torch.randn(1, 3, 224, 224)

									flops, params = profile(model, inputs=(input,))

									print(flops)

									print(params)

或者

				?

									from torchvision.models import resnet50

									from thop import profile

									# model = resnet50()

									checkpoints = '模型path'

									model = torch.load(checkpoints)

									model_name = 'yolov3 cut asff'

									input = torch.randn(1, 3, 224, 224)

									flops, params = profile(model, inputs=(input, ),verbose=True)

									print("%s | %.2f | %.2f" % (model_name, params / (1000 ** 2), flops / (1000 ** 3)))#这里除以1000的平方，是为了化成M的单位，

注意:输入必须是四维的

提高输出可读性, 加入一下代码。

				?

									from thop import clever_format

									macs, params = clever_format([flops, params], "%.3f")

方法4 torchstat

				?

									from torchstat import stat

									from torchvision.models import resnet50, resnet101, resnet152, resnext101_32x8d

									model = resnet50()

									stat(model, (3, 224, 224))  #  (3,224,224)表示输入图片的尺寸

使用torchstat这个库来查看网络模型的一些信息，包括总的参数量params、MAdd、显卡内存占用量和FLOPs等。需要安装torchstat:

				?

									pip install torchstat

方法5 ptflops

作用：计算模型参数总量和模型计算量

安装方法：pip install ptflops

或者

				?

									pip install --upgrade git+https://github.com/sovrasov/flops-counter.pytorch.git

使用方法

				?

									import torchvision.models as models

									import torch

									from ptflops import get_model_complexity_info

									with torch.cuda.device(0):

									  net = models.resnet18()

									  flops, params = get_model_complexity_info(net, (3, 224, 224), as_strings=True, print_per_layer_stat=True) #不用写batch_size大小，默认batch_size=1

									  print('Flops:  ' + flops)

									  print('Params: ' + params)

或者

				?

									from torchvision.models import resnet50

									import torch

									import torchvision.models as models

									# import torch

									from ptflops import get_model_complexity_info

									# model = models.resnet50()   #调用官方的模型，

									checkpoints = '自己模型的path'

									model = torch.load(checkpoints)

									model_name = 'yolov3 cut'

									flops, params = get_model_complexity_info(model, (3,320,320),as_strings=True,print_per_layer_stat=True)

									print("%s |%s |%s" % (model_name,flops,params))

注意，这里输入一定是要tuple类型，且不需要输入batch，直接输入输入通道数量与尺寸，如（3,320,320） 320为网络输入尺寸。

输出为网络模型的总参数量（单位M，即百万）与计算量（单位G，即十亿）

方法6 torchsummary

安装：pip install torchsummary

使用方法：

				?

									from torchsummary import summary

									...

									summary(your_model, input_size=(channels, H, W))

作用：

1、每一层的类型、shape 和参数量

2、模型整体的参数量

3、模型大小，和 fp/bp 一次需要的内存大小，可以用来估计最佳 batch_size

补充：pytorch计算模型算力与参数大小

ptflops介绍

官方链接

这个脚本设计用于计算卷积神经网络中乘法-加法操作的理论数量。它还可以计算参数的数量和打印给定网络的每层计算成本。

支持layer：Conv1d/2d/3d，ConvTranspose2d，BatchNorm1d/2d/3d，激活(ReLU, PReLU, ELU, ReLU6, LeakyReLU)，Linear，Upsample，Poolings (AvgPool1d/2d/3d、MaxPool1d/2d/3d、adaptive ones)

安装要求:Pytorch >= 0.4.1, torchvision >= 0.2.1

get_model_complexity_info()

get_model_complexity_info是ptflops下的一个方法，可以计算出网络的算力与模型参数大小，并且可以输出每层的算力消耗。

栗子

以输出Mobilenet_v2算力信息为例：

				?

									from ptflops import get_model_complexity_info

									from torchvision import models

									net = models.mobilenet_v2()

									ops, params = get_model_complexity_info(net, (3, 224, 224), as_strings=True,                                        print_per_layer_stat=True, verbose=True)