A PyTorch implementation of MobileNetV3

This is a PyTorch implementation of MobileNetV3 architecture as described in the paper Searching for MobileNetV3.

Some details may be different from the original paper, welcome to discuss and help me figure it out.

[NEW] The pretrained model of small version mobilenet-v3 is online.

Training & Accuracy

In progress ...

MobileNetV3 large

	Madds	Parameters	Top1-acc	Pretrained Model
Offical 1.0	219 M	5.4 M	75.2%	-
Offical 0.75	155 M	4 M	73.3%	-
Ours 1.0	- M	5.08 M	-	-
Ours 0.75	- M	3.69 M	-	-

MobileNetV3 small

	Madds	Parameters	Top1-acc	Pretrained Model
Offical 1.0	66 M	2.9 M	67.4%	-
Offical 0.75	44 M	2.4 M	65.4%	-
Ours 1.0	68 M	3.11 M	67.218%	[google drive]
Ours 0.75	- M	2.47 M	-	-

Usage

Pretrained models are still training ...

    # pytorch 1.0.1
    # large
    net_large = mobilenetv3(mode='large')
    # small
    net_small = mobilenetv3(mode='small')
    state_dict = torch.load('mobilenetv3_small_67.218.pth.tar')
    net_small.load_state_dict(state_dict)

Data Pre-processing

I used the following code for data pre-processing on ImageNet:

normalize = transforms.Normalize(mean=[0.485, 0.456, 0.406],
                                 std=[0.229, 0.224, 0.225])

input_size = 224

train_loader = torch.utils.data.DataLoader(
    datasets.ImageFolder(
    traindir, transforms.Compose([
        transforms.RandomResizedCrop(input_size), 
        transforms.RandomHorizontalFlip(),
        transforms.ToTensor(),
        normalize,
    ])), 
    batch_size=batch_size, shuffle=True,
    num_workers=n_worker, pin_memory=True)

val_loader = torch.utils.data.DataLoader(
    datasets.ImageFolder(valdir, transforms.Compose([
        transforms.Resize(int(input_size/0.875)),
        transforms.CenterCrop(input_size),
        transforms.ToTensor(),
        normalize,
    ])),
    batch_size=batch_size, shuffle=False,
    num_workers=n_worker, pin_memory=True)