Need to make input image more SKETCHED

[99-NinetyNine]

To do this:

1. find connected components
2. remove them (10% to 90%) as augmentation process

Hypothesis:
Removing more edges and training the model to predict not in the graph pattern, there is possibility that it learns. But it might associate strongly with color, such as a pattern of kind this would always be mapped to color; well that's for next research :).

So, after making a dataset, train the model too. The dataset pre-processing script is available here.
https://www.kaggle.com/code/kaglekuggul/to-make-datasets-bigger-and-pre-porcesse/edit

Need to improvise the script.