安卓>> CIFAR-10-transfer-learning>> 返回
项目作者： PrzemekPobrotyn

项目描述 ：
Transfer learning on CIFAR-10 using ResNet50 in Keras.
高级语言： Jupyter Notebook
项目地址: git://github.com/PrzemekPobrotyn/CIFAR-10-transfer-learning.git
创建时间： 2018-10-04T15:21:20Z
项目社区：https://github.com/PrzemekPobrotyn/CIFAR-10-transfer-learning
开源协议：MIT License
下载


CIFAR-10 TRANSFER LEARNING

This repository contains a transfer learning exercises on CIFAR-10 done in Keras.

We provided utilities to download, extract and visualise the data.

A number of models are fitted:

  • baseline: HOG features + linear SVM
  • SVM on top of CNN codes extracted using ResNet50 pretrained on ImageNet
  • Fine tuning of ResNet50 (with discussion of suitability of Keras BN layer to fine tuning task)
  • Fine tuning with data augmentation

Both development and training were conducted on Google Colab.
If you want to recreate Google Colab environment locally, pip install -r requirements.txt in your virtualenv.

In custom_resnet directory you can find code needed to perform transfer learning with ResNet50 in Keras. It was adapted from keras-applications by Keras Team and keras-resnet by Broad Institute.