在 Tensorflow 中使用迁移学习训练 Mobilnet V2

Question

各位，

如何使用 mobilenet v2 进行 transfer/domain 学习？我使用 mobilnet v1 训练代码来训练 mobilenet v1，但是在模型存储库或 tensorflow 存储库中是否有用于训练 v2 的书面代码？

https://github.com/tensorflow/models/tree/master/research

https://github.com/tensorflow/tensorflow/tree/master/tensorflow/examples/image_retraining

Answer 1

This has the mobilenet v2 tfslim modules, as well as the checkpoint files to restore weights already trained by the tensorflow people. You can use this example to retrain the model, or you can take a simpler approach with this tutorial. If you're looking for portability, might want to look at TFLite, which would allow you to take any tensorflow model and easily optimize it for mobile with the build tools at tensorflow head: https://github.com/tensorflow/tensorflow/tree/master/tensorflow/python/tools。

祝你好运

Answer 2

@Asher Mancinelli，实际上，这段代码正是我需要的： https://raw.githubusercontent.com/tensorflow/hub/r0.1/examples/image_retraining/retrain.py