说明文档

Guide

Tensorflow based VGG16 and AlexNet training, validation and test pipeline.

A batch generator with a shuffle in each epoch.

Friendly command line interface and log management.

With Kaggle Digit Recognition as Example.

Just try with ease.

运行环境

默认调用GPU，要求train.csv与test.csv与python文件处于同一目录.

训练

默认按train.csv样本顺序取前80%作为训练集

AlexNet

python alexnet_train.py --lr 0.0001 --weightDecay 0.1 --maxIter 20000 --batchSize 30 --trainRatio 0.8

VGG16

python vgg16_train.py --lr 0.0001 --weightDecay 0.1 --maxIter 20000 --batchSize 30

验证

默认按train.csv样本顺序取后20%作为验证集

AlexNet

python alexnet_train.py --isVal --modelPath /path/to/your/model.cpkt

VGG16

python vgg16_train.py --isVal --modelPath /path/to/your/model.cpkt

测试

将在当前目录生成submission.csv文件

AlexNet

python alexnet_train.py --isTest --modelPath /path/to/your/model.cpkt

VGG16

python vgg16_train.py --isTest --modelPath /path/to/your/model.cpkt

more

python alexnet_train.py --help
python vgg16_train.py --help