Skip to content

Format of h5 augmented training data

Chris Churas edited this page Mar 1, 2018 · 6 revisions

This page describes the format of the HDF5 .h5 augmented training data files generated by PreprocessTrainingData.m. This is a technical document that for anyone that wants to know the inner workings.

PreprocessTrainingData.m generates 16 .h5 files each

Format of individual .h5 file

The image data resides in the data Dataset and the mask (or features to segment) reside in the label Dataset. For the augmented images there is NO label dataset.

TODO: Add figure showing how images are stored in data and label datasets

Stacking

V1 through V8

Images are stacked from 0 to n-1

V9 through V16

Images are stacked in opposite order n-1 to 0

Image manipulations

V1 & V9

No change

V2 & V10

Images are flipped on horizontal access

V3 & V11

Images are flipped on vertical access

V4 & V12

Images are rotated 90 degrees counter clockwise

V5 & V13

Images are rotated 90 degrees clockwise

V6 & V14

Images are rotated 90 degrees counter clockwise and flipped on vertical access

V7 & V15

Images are rotated 90 degrees counter clockwise and flipped on horizontal access

V8 & V16

Images are rotated 180 degrees counter clockwise

Clone this wiki locally