Benchmark - SENet/SKNet/ResNeSt
Convert
The senet/sknet/resnest pretraining model can be transformed into zcls by script
$ cd /path/to/ZCls
$ python tools/converters/moskomule_se_resnet_to_zcls_se_resnet.py
$ python tools/converters/syt2_sknet_to_zcls_sknet.py
$ python tools/converters/official_resnest_to_zcls_resnest.py
Benchmark
| arch | framework | top1 | top5 | input_size | dataset | params_size/MB |
gflops |
cpu_infer/s |
|---|---|---|---|---|---|---|---|---|
| se_resnet50 | zcls | 75.500 | 92.720 | 224x224 | imagenet | 107.086 | 8.239 | 0.073 |
| sknet50 | zcls | 78.343 | 93.944 | 224x224 | imagenet | 104.885 | 9.000 | 0.143 |
| resnest50_fast_2s1x64d | zcls | 80.394 | 95.254 | 224x224 | imagenet | 104.840 | 8.719 | 0.092 |
| resnest50_fast_2s1x64d | official | 80.380 | 95.252 | 224x224 | imagenet | 104.840 | 8.716 | 0.083 |
| resnest50 | zcls | 80.748 | 95.399 | 224x224 | imagenet | 104.840 | 10.805 | 0.435 |
| resnest50 | official | 80.742 | 95.397 | 224x224 | imagenet | 104.840 | 10.801 | 0.424 |
| resnest101 | zcls | 81.942 | 95.931 | 224x224 | imagenet | 184.155 | 20.496 | 0.667 |
| resnest101 | official | 81.938 | 95.927 | 224x224 | imagenet | 184.155 | 20.491 | 0.623 |
| resnest200 | zcls | 82.262 | 96.101 | 224x224 | imagenet | 267.798 | 34.991 | 0.344 |
| resnest200 | official | 82.262 | 96.103 | 224x224 | imagenet | 267.798 | 34.981 | 0.350 |
| resnest269 | zcls | 78.659 | 94.176 | 224x224 | imagenet | 423.162 | 45.083 | 0.454 |
| resnest269 | official | 78.659 | 94.176 | 224x224 | imagenet | 423.162 | 45.070 | 0.476 |
CPU_INFO: Intel(R) Core(TM) i7-7700HQ CPU @ 2.80GHz