统计官方模型的参数量和计算量

Model	Params(M)	FLOPs(G)
alexnet	61.10	0.71
densenet121	7.98	2.90
densenet161	28.68	7.85
densenet169	14.15	3.44
densenet201	20.01	4.39
googlenet	6.62	1.51
inception_v3	23.83	5.75
mnasnet0_5	2.22	0.12
mnasnet0_75	3.17	0.23
mnasnet1_0	4.38	0.34
mnasnet1_3	6.28	0.56
mobilenet_v2	3.50	0.33
mobilenet_v3_large	5.48	0.23
mobilenet_v3_small	2.54	0.06
resnet101	44.55	7.87
resnet152	60.19	11.60
resnet18	11.69	1.82
resnet34	21.80	3.68
resnet50	25.56	4.13
resnext101_32x8d	88.79	16.54
resnext50_32x4d	25.03	4.29
shufflenet_v2_x0_5	1.37	0.04
shufflenet_v2_x1_0	2.28	0.15
shufflenet_v2_x1_5	3.50	0.31
shufflenet_v2_x2_0	7.39	0.60
squeezenet1_0	1.25	0.82
squeezenet1_1	1.24	0.35
vgg11	132.86	7.61
vgg11_bn	132.87	7.64
vgg13	133.05	11.31
vgg13_bn	133.05	11.36
vgg16	138.36	15.47
vgg16_bn	138.37	15.52
vgg19	143.67	19.63
vgg19_bn	143.68	19.69
wide_resnet101_2	126.89	22.84
wide_resnet50_2	68.88	11.46

import torch
from torchvision import models

model_names = sorted(
    name
    for name in models.__dict__
    if name.islower()
    and not name.startswith("__")  # and "inception" in name
    and callable(models.__dict__[name])
)

device = "cpu"


for name in model_names:
    print("======================="+name+"=========================")
    model = models.__dict__[name]().to(device)
    dsize = (1, 3, 224, 224)
    size_ = (3, 224, 224)
    if "inception" in name:
        dsize = (1, 3, 299, 299)
        size_ = (3, 224, 224)

    '''
    # thop，功能单一，不推荐
    from thop.profile import profile
    inputs = torch.randn(dsize).to(device)
    total_ops, total_params = profile(model, (inputs,), verbose=False)
    print(
        "%s | %.2f | %.2f" % (name, total_params / (1000 ** 2), total_ops / (1000 ** 3))
    )
    '''

    '''
    # 直接测量参数量，不能统计计算量。
    total_params = sum(param.numel() for param in model.parameters())
    print(total_params / (1000 ** 2))
    '''

    # torchstat，打印内容最详细，但没打印层名，且支持某些层的种类有限（https://github.com/Swall0w/torchstat/blob/master/detail.md）。对于包含这些层的网络，计算结果比实际偏低。测量的是GPU模型。
    from torchstat import stat
    stat(model.to(device), size_)
    
    # torchinfo。框架比较新，还在更新中，各层的名称比较明确，但打印信息不如torchstat全面。测量的是CPU模型。
    from torchinfo import summary
    summary(model, input_size=dsize)