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)