-1

In 《MnasNet: Platform-Aware Neural Architecture Search for Mobile》, the author said that they have plugged their learned model architecture into the open-source TensorFlow Object Detection framework, as a new feature extractor. But I searched the model of Tensorflow in Github and couldn't find the code. Does anyone know the position of the training codes of MnasNet in tf?

user2906838
  • 1,178
  • 9
  • 20
alvin_zhu
  • 1

1 Answers1

0

Actually there's an implementation in tensorflow (in tf.keras) in github (non-official implementation)

Íñigo Alonso Ruiz
  • 1
  • https://github.com/Shathe/MNasNet-Keras-Tensorflow/blob/master/Mnasnet.py – Íñigo Alonso Ruiz Aug 16 '18 at 08:28
  • Hi. As links are perishable over time, please could you add the most relevant parts of what you linked to in your answer, so that if the links are ever deleted this answer is still valuable? – James Whiteley Aug 16 '18 at 08:36
  • @ÍñigoAlonsoRuiz Thanks, I've read the codes of MnasNet. But the mansnet.py and train.py just show the formation of models with convs & bns. I'm curious about how they train to get the model with reinforcement learning and how to use latency in rewards. It looks like Google's open-source Enas, but has a lot of difference. – alvin_zhu Aug 16 '18 at 08:53