Benchmark - ResNet/ResNeXt

Convert

The torchvision pretraining model can be transformed into zcls by script

$ cd /path/to/ZCls
$ python tools/converters/torchvision_resnet_to_zcls_resnet.py

Benchmark

arch	framework	top1	top5	input_size	dataset	params_size/MB	gflops	cpu_infer/s
resnet18	zcls	69.224	88.808	224x224	imagenet	44.592	3.638	0.030
resnet18	torchvision	69.222	88.808	224x224	imagenet	44.592	3.638	0.032
resnet34	zcls	72.821	91.071	224x224	imagenet	83.152	7.343	0.060
resnet34	torchvision	72.817	91.073	224x224	imagenet	83.152	7.343	0.055
resnet50	zcls	75.692	92.768	224x224	imagenet	97.492M	8.223	0.082
resnet50	torchvision	75.690	92.770	224x224	imagenet	97.492	8.223	0.078
resnet101	zcls	76.965	93.526	224x224	imagenet	169.942	15.668	0.145
resnet101	torchvision	76.967	93.528	224x224	imagenet	169.942	15.668	0.144
resnet152	zcls	78.141	93.946	224x224	imagenet	229.617	23.118	0.208
resnet152	torchvision	78.139	93.946	224x224	imagenet	229.617	23.118	0.210
resnext50_32x4d	zcls	77.353	93.610	224x224	imagenet	95.478	8.519	0.106
resnext50_32x4d	torchvision	77.353	93.612	224x224	imagenet	95.478	8.519	0.103
resnext101_32x8d	zcls	79.075	94.540	224x224	imagenet	338.712	32.953	0.311
resnext101_32x8d	torchvision	79.063	94.538	224x224	imagenet	338.712	32.953	0.298

CPU_INFO: Intel(R) Core(TM) i7-7700HQ CPU @ 2.80GHz