MNasNet

Keras (Tensorflow) Implementation of MNasNet and an example for training and evaluating it on the MNIST dataset. Check also the eager execution implementation.

A Tensorflow Keras implementation (Graph and eager execution) of Mnasnet: MnasNet: Platform-Aware Neural Architecture Search for Mobile.

Requirement

  • Python 2.7+
  • Tensorflow-gpu 1.10

Train it

Train the MNasNet model on the MNIST dataset! just execute:

python train.py

For checking and inspecting the Mnasnet model described in the paper, execute:

python Mnasnet.py

Train it with eager execution

Train the MNasNet (eager) model on the MNIST dataset! just execute:

python train_eager.py

The eager execution implementation also outputs logs on Tensorboard. For its visualization:

tensorboard --logdir=train_log:./logs/train, test_log:./logs/test

MnasNet for... Semantic Segmentation!

In this other repository, FC-Mnasnet I added a decoder to the MnasNet architecture in order to turn it into a semantic segmentation model.

mnasnet

GitHub