Benchmark - MobileNet
Convert
The torchvision pretraining model can be transformed into zcls by script
$ cd /path/to/ZCls
$ python tools/converters/torchvision_mobilenet_to_zcls_mobilenet.py
Benchmark
arch | framework | top1 | top5 | input_size | dataset | params_size/MB |
gflops |
cpu_infer/s |
---|---|---|---|---|---|---|---|---|
mobilenet_v2 | zcls | 71.429 | 90.151 | 224x224 | imagenet | 13.370 | 0.628 | 0.027 |
mobilenet_v2 | torchvision | 71.429 | 90.151 | 224x224 | imagenet | 13.370 | 0.628 | 0.025 |
mnasnet0_5 | zcls | 66.961 | 86.946 | 224x224 | imagenet | 8.463 | 0.221 | 0.017 |
mnasnet0_5 | torchvision | 66.961 | 86.946 | 224x224 | imagenet | 8.463 | 0.221 | 0.015 |
mnasnet1_0 | zcls | 73.249 | 91.235 | 224x224 | imagenet | 16.721 | 0.651 | 0.021 |
mnasnet1_0 | torchvision | 73.249 | 91.235 | 224x224 | imagenet | 16.721 | 0.651 | 0.023 |
mobilenet_v3_large | torchvision | 65.157 | 86.014 | 224x224 | imagenet | 20.916 | 0.451 | 0.131 |
mobilenet_v3_small | torchvision | 70.909 | 89.765 | 224x224 | imagenet | 9.700 | 0.119 | 0.097 |
CPU_INFO: Intel(R) Core(TM) i7-7700HQ CPU @ 2.80GHz