splinter21 / mobilenetv3.pytorch

73.2% MobileNetV3-Large and 67.1% MobileNetV3-Small model on ImageNet

PyTorch Implementation of MobileNet V3

Reproduction of MobileNet V3 architecture as described in Searching for MobileNetV3 by Andrew Howard, Mark Sandler, Grace Chu, Liang-Chieh Chen, Bo Chen, Mingxing Tan, Weijun Wang, Yukun Zhu, Ruoming Pang, Vijay Vasudevan, Quoc V. Le, Hartwig Adam on ILSVRC2012 benchmark with PyTorch framework.

Requirements

Dataset

Download the ImageNet dataset and move validation images to labeled subfolders. To do this, you can use the following script: https://raw.githubusercontent.com/soumith/imagenetloader.torch/master/valprep.sh

Models

Architecture	# Parameters	MFLOPs	Top-1 / Top-5 Accuracy (%)
MobileNetV3-Large	5.145M	245.58	73.152 / 91.102
MobileNetV2 1.0	3.504M	300.79	72.192 / 90.534
MobileNetV3-Small	3.112M	57.08	67.102 / 86.374
MobileNetV2 0.35	1.677M	59.29	60.092 / 82.172

Note: The implemented architecture follows Table 1 and 2 in the paper, yet architectural details are vaguely described, rendering mismatches of both parameters and complexity.

from mobilenetv3 import mobilenetv3_large, mobilenetv3_small

net_large = mobilenetv3_large()
net_small = mobilenetv3_small()

net_large.load_state_dict(torch.load('pretrained/mobilenetv3-large-657e7b3d.pth'))
net_small.load_state_dict(torch.load('pretrained/mobilenetv3-small-c7eb32fe.pth'))

Citation

@ARTICLE{2019arXiv190502244H,
       author = {{Howard}, Andrew and {Sandler}, Mark and {Chu}, Grace and
         {Chen}, Liang-Chieh and {Chen}, Bo and {Tan}, Mingxing and
         {Wang}, Weijun and {Zhu}, Yukun and {Pang}, Ruoming and
         {Vasudevan}, Vijay and {Le}, Quoc V. and {Adam}, Hartwig},
        title = "{Searching for MobileNetV3}",
      journal = {arXiv e-prints},
     keywords = {Computer Science - Computer Vision and Pattern Recognition},
         year = "2019",
        month = "May",
          eid = {arXiv:1905.02244},
        pages = {arXiv:1905.02244},
archivePrefix = {arXiv},
       eprint = {1905.02244},
 primaryClass = {cs.CV},
       adsurl = {https://ui.adsabs.harvard.edu/abs/2019arXiv190502244H},
      adsnote = {Provided by the SAO/NASA Astrophysics Data System}
}