Benchmark - ShuffleNet
Convert
The official shufflenet_v1/torchvision shufflenet_v2 pretraining model can be transformed into zcls by script
$ cd /path/to/ZCls
$ python tools/converters/official_shufflenet_v1_to_zcls_shufflenet_v1.py
$ python tools/converters/torchvision_shufflenet_to_zcls_shufflenet.py
Benchmark
arch | framework | top1 | top5 | input_size | dataset | params_size/MB |
gflops |
cpu_infer/s |
---|---|---|---|---|---|---|---|---|
shufflenet_v1_3g1x | zcls | / | / | 224x224 | imagenet | 7.113 | 0.287 | 0.019 |
shufflenet_v1_8g1x | zcls | / | / | 224x224 | imagenet | 9.284 | 0.296 | 0.022 |
shufflenet_v2_x0_5 | zcls | 59.729 | 81.226 | 224x224 | imagenet | 5.210 | 0.085 | 0.013 |
shufflenet_v2_x0_5 | torchvision | 59.729 | 81.230 | 224x224 | imagenet | 5.214 | 0.085 | 0.012 |
shufflenet_v2_x1_0 | zcls | 68.942 | 88.218 | 224x224 | imagenet | 8.693 | 0.300 | 0.018 |
shufflenet_v2_x1_0 | torchvision | 68.940 | 88.218 | 224x224 | imagenet | 8.692 | 0.298 | 0.018 |
CPU_INFO: Intel(R) Core(TM) i7-7700HQ CPU @ 2.80GHz