PyTorch中FLOPs计算问题

2022-02-05 13:32:46 阅读：245 来源： 互联网

标签：layer nn self stride PyTorch 计算 FLOPs norm out

最近看了很多关于FLOPs计算的实现方法，也自己尝试了一些方法，发现最好用的还是PyTorch中的thop库（代码如下）：

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model = 模型的名字().to(device)
inputs = torch.randn(1,3,512,1024)   ####(360,640)
inputs=inputs.to(device)
macs, params = profile(model,inputs=(inputs,))   ##verbose=False
print('The number of MACs is %s'%(macs/1e9))   ##### MB
print('The number of params is %s'%(params/1e6))   ##### MB

实现起来确实很简单，那么问题来了，这里面算出来的macs到底是MACs还是FLOPs呢？先说我自己探索得到的结论，这里计算出的macs其实就是FLOPs（每秒钟浮点运算次数），前提是：不计算卷积层的bias，原因如下：

自己手动计算ResNet18的FLOPs，对于512*1024*3的输入尺寸。

（1）ResNet18的代码：

class BasicBlock(nn.Module):
    expansion = 1

    def __init__(self, inplanes, planes, r=1, stride=1, downsample=None, norm_layer=nn.BatchNorm2d):
        super(BasicBlock, self).__init__()
        # Both self.conv1 and self.downsample layers downsample the input when stride != 1
        self.conv1 = conv3x3(inplanes, planes, stride)
        self.bn1 = norm_layer(planes)
        self.relu = nn.ReLU(inplace=True)
        self.conv2 = conv3x3(planes, planes, dilation=r)
        self.bn2 = norm_layer(planes)
        self.downsample = downsample
        self.stride = stride

    def forward(self, x):
        identity = x

        out = self.conv1(x)
        out = self.bn1(out)
        out = self.relu(out)

        out = self.conv2(out)
        out = self.bn2(out)

        if self.downsample is not None:
            identity = self.downsample(x)

        out += identity
        out = self.relu(out)
        return out


class ResNet18(nn.Module):

    def __init__(self, block=BasicBlock, layers=[2,2,2,2], zero_init_residual=False, norm_layer=nn.BatchNorm2d):
        super(ResNet18, self).__init__()
        self.inplanes = 64
        self.conv1 = nn.Conv2d(3, 64, kernel_size=7, stride=2, padding=3,
                               bias=False)
        self.bn1 = norm_layer(64)
        self.relu = nn.ReLU(inplace=True)
        self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
        self.layer1 = self._make_layer(block, 64, layers[0], r=2, norm_layer=norm_layer)
        self.layer2 = self._make_layer(block, 128, layers[1], stride=2, r=2, norm_layer=norm_layer)
        self.layer3 = self._make_layer(block, 256, layers[2], stride=2, r=2, norm_layer=norm_layer)
        self.layer4 = self._make_layer(block, 512, layers[3], stride=2, r=2, norm_layer=norm_layer)

        for m in self.modules():
            if isinstance(m, nn.Conv2d):
                nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
            elif isinstance(m, nn.BatchNorm2d):
                nn.init.constant_(m.weight, 1)
                nn.init.constant_(m.bias, 0)

        # Zero-initialize the last BN in each residual branch,
        # so that the residual branch starts with zeros, and each residual block behaves like an identity.
        # This improves the model by 0.2~0.3% according to https://arxiv.org/abs/1706.02677
        if zero_init_residual:
            for m in self.modules():
                if isinstance(m, Bottleneck):
                    nn.init.constant_(m.bn3.weight, 0)
                elif isinstance(m, BasicBlock):
                    nn.init.constant_(m.bn2.weight, 0)

    def _make_layer(self, block, planes, blocks, stride=1, r=1, norm_layer=nn.BatchNorm2d):
        downsample = None
        if stride != 1 or self.inplanes != planes * block.expansion:
            downsample = nn.Sequential(
                conv1x1(self.inplanes, planes * block.expansion, stride),
                norm_layer(planes * block.expansion),
            )

        layers = []
        layers.append(block(self.inplanes, planes, r, stride, downsample))
        self.inplanes = planes * block.expansion
        for _ in range(1, blocks):
            layers.append(block(self.inplanes, planes))

        return nn.Sequential(*layers)

    def forward(self, x):
        x = self.conv1(x)
        x = self.bn1(x)
        x = self.relu(x)
        x = self.maxpool(x)

        x = self.layer1(x)
        x = self.layer2(x)
        x_downsampling_8 = x       ###(h/8,128)
        x = self.layer3(x)
        x_downsampling_16 = x      ###(h/16,256)
        x = self.layer4(x)         ###(h/32,512)

        return x, x_downsampling_8, x_downsampling_16

（2）用profile函数计算得到的macs值为 19.0GB

（3）自己手动计算FLOPs ≈ 20GB

因此，在不统计卷积层bias计算次数的前提下，profile函数计算得到的macs值其实就是FLOPs。

（PS：个人理解，欢迎批评纠正）

标签：layer,nn,self,stride,PyTorch,计算,FLOPs,norm,out
来源： https://blog.csdn.net/qq_38596190/article/details/122789694

本站声明： 1. iCode9 技术分享网（下文简称本站）提供的所有内容，仅供技术学习、探讨和分享；
2. 关于本站的所有留言、评论、转载及引用，纯属内容发起人的个人观点，与本站观点和立场无关；
3. 关于本站的所有言论和文字，纯属内容发起人的个人观点，与本站观点和立场无关；
4. 本站文章均是网友提供，不完全保证技术分享内容的完整性、准确性、时效性、风险性和版权归属；如您发现该文章侵犯了您的权益，可联系我们第一时间进行删除；
5. 本站为非盈利性的个人网站，所有内容不会用来进行牟利，也不会利用任何形式的广告来间接获益，纯粹是为了广大技术爱好者提供技术内容和技术思想的分享性交流网站。

ICode9

PyTorch中FLOPs计算问题