I'm trying to get familiar with modelling a CNN for an image detection project. Right now I'm stuck with preprocessing the image training data set:
I have labeled 40 images with labelbox. Now I want to go on for image Data Augmentation. That is I want to rotate, flip etc. all labeled images in my data set to get more training datas to feed in my CNN. I want to do it with the "data_aug"-package from "https://augmentationlib.paperspace.com/index.html" as it does Data Augmentation with labeled images (that is images with the boundary boxes). The problem is I don't know how to install this modul as the command pip or apt get doesn't run in the cell.
Which command do I need? How can I install such a package? Or is there any much simpler way to do Data Augmentation with labeled images? Thanks in advance!
I had a look around and it seems that "package" isn't so much a package, as it is a project with a Github repo available. So it isn't available via standard package builders; you'll have to manually download the data_aug guys' code. Luckily it's not hard.
To start, just clone their repo: https://github.com/Paperspace/DataAugmentationForObjectDetection
All the code is now available to you. they have a tutorial on how to use it over here: https://blog.paperspace.com/data-augmentation-for-bounding-boxes/