MNASNet 1.0

Last updated on Feb 12, 2021

MNASNet 1.0

Parameters 4 Million
FLOPs 325 Million
File Size 16.92 MB
Training Data ImageNet
Training Resources 8x NVIDIA V100 GPUs
Training Time

Training Techniques Weight Decay, SGD with Momentum
Architecture 1x1 Convolution, Batch Normalization, Convolution, Depthwise Separable Convolution, Dropout, Global Average Pooling, Inverted Residual Block, Residual Connection, ReLU, Max Pooling, Softmax, Squeeze-and-Excitation Block
ID mnasnet1_0
LR 0.1
Epochs 90
LR Gamma 0.1
Momentum 0.9
Batch Size 32
LR Step Size 30
Weight Decay 0.0001
SHOW MORE
SHOW LESS
README.md

Summary

MnasNet is a type of convolutional neural network optimized for mobile devices that is discovered through mobile neural architecture search, which explicitly incorporates model latency into the main objective so that the search can identify a model that achieves a good trade-off between accuracy and latency. The main building block is an inverted residual block (from MobileNetV2).

How do I load this model?

To load a pretrained model:

import torchvision.models as models
mnasnet = models.mnasnet1_0(pretrained=True)

Replace the model name with the variant you want to use, e.g. mnasnet1_0. You can find the IDs in the model summaries at the top of this page.

To evaluate the model, use the image classification recipes from the library.

python train.py --test-only --model='<model_name>'

How do I train this model?

You can follow the torchvision recipe on GitHub for training a new model afresh.

Citation

@article{DBLP:journals/corr/abs-1807-11626,
  author    = {Mingxing Tan and
               Bo Chen and
               Ruoming Pang and
               Vijay Vasudevan and
               Quoc V. Le},
  title     = {MnasNet: Platform-Aware Neural Architecture Search for Mobile},
  journal   = {CoRR},
  volume    = {abs/1807.11626},
  year      = {2018},
  url       = {http://arxiv.org/abs/1807.11626},
  archivePrefix = {arXiv},
  eprint    = {1807.11626},
  timestamp = {Mon, 13 Aug 2018 16:46:25 +0200},
  biburl    = {https://dblp.org/rec/journals/corr/abs-1807-11626.bib},
  bibsource = {dblp computer science bibliography, https://dblp.org}
}

Results

Image Classification on ImageNet

Image Classification
BENCHMARK MODEL METRIC NAME METRIC VALUE GLOBAL RANK
ImageNet MNASNet 1.0 Top 1 Accuracy 73.51% # 247
Top 5 Accuracy 91.54% # 247