精度比resnetv2高2个点
https://github.com/zhanghang1989/ResNeSt
1070 512 batch size 只能是4 100ms
net=resnest50().cuda()
import time
net.eval()
x = torch.randn(4, 3, 512, 512).cuda()
# x = torch.randn(64,224,7,7).cuda()
for i in range(10):
start = time.time()
out = net(x)
print(time.time() - start, out.size())