ImgLibMaker Manual

1. Introduction

This is a simple program to zip image file(.jpg) to .plk file with Python pickle library. The python version is 3.6.4.

If you are working on Image classify/object detection area and you do not want to use those standard image library(mnist/cifar .etc), this program is suitable for you.

2. How to use

The main program is ImageHandle.py and there are four input arguments for the program:

• --Path_Dir
• --reshape
• --img_num_each_file
• --test_num_each_class

--Path_Dir is the direction of your images stored, your image recommanded to store in one folder like this:

So for mine, the --Path_Dir is "D:\xuexiziliao\Proj\VS PROJ\P1\RGB_IMG" shown on the up dialog. In this folder have several items, contains three folders and other documents.

It is very important that you can only contains folders which contains your images, for example, mine have three folders: P-0000,P-0001,P-0002. If you want to make your own image library, the folder name must follow the style: P-XXXX.

So what's the meaning? I have three image classes, first class is 'human hands', those pictures are all contains in P-0000, second class is 'mouse' which contains in P-0001 and the last is 'cell phone' which contains in P-0002. If you run this program, it will generate a (or several) .plk file(s) which contains the raw images and their labels(classes), so P-0000 will have label '0', P-0001 will have label '1' etc.

So do not contains any folder which is not for zipping your images. Again, the folder name must be P-XXXX.

--reshape means reshape all the images to the same size, it takes a tuple, the default is (320,240).

--img_num_each_file means how many pictures each .plk file contains. the default is 500, if your image number exceed this parameter, it will generate another .plk that increase its index automaticly

--test_num_each_class means how many images for test for each class. Due to this program is for neural network training, so the for ordinary, we should have a training image set and a test image set, this parameter is for this porpuse, if you set it to 100, it will take out 100 images from each folder names 'P-XXXX' and generate a Test_DataX.plk file. The default number is 0.

You can use following program to load .plk file:

  testfile=open(os.path.join(Path_Dir,'Test_Data0.pkl'),'rb')
  image=pickle.load(testfile)
  label=pickle.load(testfile)
  testfile2.close

The pickle file contains two dictionary one names 'image', the other names 'label', note that the program reshape the image in one row, so if you want to restore the image, you should reshape by yourself.