[FEATURE] ImageDataGenerator for Hub

[DebadityaPal]

🚨🚨 Feature Request

• [ ] Related to an existing Issue
• [x] A new implementation (Improvement, Extension)

Is your feature request related to a problem?

Data Augmentation is an important step while working with small datasets in Computer Vision. An incentive for not using Hub and storing data on local machines right now is the data augmentation pipeline.

With Keras, a developer can just set up an ImageDataGenerator and then call the flow_from_directory method.

For Hub, this step is very complex. The data must first be fetched and then the dev has to explicitly code all the augmentations in the data generator.

If your feature will improve HUB

This feature will elevate hub in terms of ease of use. Thus, more people might choose hub over storing data locally due to the aforementioned reason.

[FayazRahman]

Will it be okay to use scipy as a dependency for applying transformations?

[dhiganthrao]

Looping @AbhinavTuli into this conversation as he can possibly provide more context on this issue.