JiahuiYu / slimmable_networks

Slimmable Networks, AutoSlim, and Beyond, ICLR 2019, and ICCV 2019
Other
914 stars 131 forks source link

top1 error = 0.99 when testing #48

Closed YoungDav closed 4 years ago

YoungDav commented 4 years ago

python==3.5.6 torch==1.1.0 torchvision==0.4.1

  1. mkdir logs/ && Download pretrain-model
  2. ln -s imagenet_datadir ./
  3. python train.py app:apps/s_mobilenet_v1_train_val.yml or python train.py app:apps/us_mobilenet_v1_train_val.yml
  4. Error log:

77.3s test 1.0 0/480: loss: 20.153, top5_error: 0.997, top1_error: 0.999 77.3s test 0.75 0/480: loss: 19.319, top5_error: 0.997, top1_error: 0.999 77.4s test 0.5 0/480: loss: 17.999, top5_error: 0.997, top1_error: 0.999 77.4s test 0.25 0/480: loss: 15.237, top5_error: 0.996, top1_error: 0.999 66.0s test 1.0 0/480: loss: 20.153, top5_error: 0.997, top1_error: 0.999 66.0s test 0.75 0/480: loss: 19.319, top5_error: 0.997, top1_error: 0.999 66.1s test 0.5 0/480: loss: 17.999, top5_error: 0.997, top1_error: 0.999 66.1s test 0.25 0/480: loss: 15.237, top5_error: 0.996, top1_error: 0.999

thanks

YoungDav commented 4 years ago

error come from dataset format no match