Note: This project is pytorch implementation of Mobilenet.
Trained on ImageNet, get Prec@1 63.550% Prec@5 85.650%. Unfortunately, due to limitation of GPU resources, training was stopped at 39 epoch. So I think it may get better results if the training could be completed. During the training, I set batch_size=256, learning_rate=0.1 which decayed every 30 epoch by 10.
python main.py -b 256 $ImagenetdirCUDA_VISIBLE_DEVICES=2,3 python main.py -b 256 $Imagenetdir| commit version | Acc@1 | Acc@5 | optimizer |
|---|---|---|---|
| d98433a | 56.852 | 79.990 | SGD |
| 3ae3b7f | 66.908 | 87.504 | SGD |
License: MIT license (MIT)