Adding image brightness data augmentation to neural network preprocessing

4 visualizaciones (últimos 30 días)
Currently the augmentedImageDatastore does not support brightness (intensity) augmentation. Could this feature be added in the future?

Respuestas (1)

Krishna
Krishna el 26 de Ag. de 2024
Hello,
I understand that you want to do data augmentation in MATLAB, and for that you need to adjust the brightness in ‘augmentedImageDatastore.’
The ‘augmentedImageDatastore’ uses the ‘imageDataAugmentor’ object where you can add augmentation directly like resizing, rotation, reflection. But it does not allow you to add brightness, contrast, hue, saturation etc. Please look at this documentation to learn more about these 2 functions,
To adjust the brightness and do more complicated data augmentations you need to create your own augmentation function. Please look at this example once.
If you look at the following example in the data augmentation section, they have created a custom data augmenter function and then used transform function on the Datastore to use fly-by data augmentation. This means when the training happens the training data (to which the transform function is applied) uses the data augmentation function to the datastore images without storing them.
Please go through the following documentation to learn more regarding transform function,
Also in their custom data augmenter function they have used the function ‘jitterColorHSV’ to change brightness and other image properties. Look at the below documenation to learn more,
Hope this helps.

Categorías

Más información sobre Deep Learning Toolbox en Help Center y File Exchange.

Productos


Versión

R2018a

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by