This is a PyTorch implementation of MobileNetV2 architecture as described in the paper Inverted Residuals and Linear Bottlenecks: Mobile Networks for Classification, Detection and Segmentation.
Note: some part of the network structure is not described clearly in the paper, so the implementation correctness is not guaranteed. Thanks for all your feedbacks.
I tried to train the model with RMSprop from scratch as described in the paper, but it does not seem to work.
I am currently training the model with SGD and keeping other hyper-parameters the same (except that I use batch size 256). I will also try fine-tuning with RMSprop from SGD checkpoint in the future.
The top-1 accuracy on the ImageNet from the paper is 71.7%. My current result is:
| Optimizer | Epoch | Top1-acc | Pretrained Model |
|---|---|---|---|
| RMSprop | - | - | - |
| SGD | 233 | 71.162% | [google drive] |
| SGD + RMSprop | TODO | TODO | TODO |
(The training is still going on since I do not have many GPUs :(, I'll update the link if I obtain better results.)
To use the pretrained model, run
from MobileNetV2 import MobileNetV2
net = MobileNetV2(n_class=1000)
state_dict = torch.load('mobilenetv2.pth.tar')
net.load_state_dict(state_dict)