GoogLeNet-TensorFlow

TensorFlow implementation of GoogLeNet.

Introduction

Focus on the structure of the project, The current code can be properly trained. Next I will continue to improve the readability of the code and the structure of the project will also improve the accuracy of the GoogLeNet network training.

To-Do

This repository will use object-oriented programming as much as possible to make the machine learning code structure clearer. So far I have implemented data loader and processing configuration classes and implemented Inception v1 network class. In addition, the current code can be visualized using tensorboard.

Usage

Data

This repository will support data in a variety of formats. Up to now it supports 102flowers dataset.

102flowers

In order to ensure the correct training, please organize the structure of the data as follows.

data
├── flowers102
│   ├── imagelabels.csv
│   └── jpg
└── imagenet_models
    └── googlenet.npy

You can download the dataset from here.

Training

All configuration of data and training can be modified in the file. Use the more readable yaml file format as the configuration file.

For more details, please refer to experiments/configs/inception_v1.yml

Simply run this script:

python ./training.py

Recommend using python virtual environment to train.

Reference

Szegedy, C., Liu, W., Jia, Y., Sermanet, P., Reed, S., Anguelov, D., et al. (2014, September 17). Going Deeper with Convolutions. arXiv.org.

Nilsback, M-E. and Zisserman, A. Automated flower classification over a large number of classes

Images

example of data

图片.png

graph

Please note that the current Tensorboard shows Graph has been beautified, but it is implemented in other ways. Because of the lower version of TensorFlow, using variable_scope when loading pre-trained weights will result in duplicate variable_scope being created. If you need to beautify Graph, please select a version of tensorflow >=1.5.0 and use uxiliary_name_scope=False