ECN>> image-tfrecord-builder>> 返回
项目作者： tinenbruno

项目描述 ：
A python tfrecord builder tool focused on images
高级语言： Python
项目地址: git://github.com/tinenbruno/image-tfrecord-builder.git
创建时间： 2018-05-31T18:29:53Z
项目社区：https://github.com/tinenbruno/image-tfrecord-builder
开源协议：MIT License
下载


Image TFRecord Builder

A python tfrecord builder tool focused on images

Usage

Dataset Setup

This tool expects your image dataset to be structured in the following way:

  1. > base_dir
  2.   > classname 1
  3.     > image 1
  4.     > image 2
  5.     > ...
  6.   > classname 2
  7.     > image 1
  8.     > image 2
  9.     > ...

Settings

You will also need to adjust the settings.py.

Setting Description Example
IMAGES_INPUT_FOLDER Absolute path for your image dataset folder /home/user/dataset
OUTPUT_FILENAME Absolute base filename for output (can be used with a remote setting as well) /home/user/output
NUMBER_OF_SHARDS Number of splits for both training and test tfrecord files 2
TRAINING_EXAMPLES_SPLIT Percentage of examples that will be used as training 0.8
SEED Seed to allow example shuffling with repeatability 123

Requirements

Using pip: pip install -r requirements.txt

Running

To run simply use python image-tfrecord-builder.py