I will do data augmentation on images

Why Augmentation Matters?

• To overcome limited data
• To improve model accuracy
• To improve model work on any environmental conditions, object rotations, dis-alignments, blurriness, scales etc

I can augment a dataset to generate images with variations. This is useful when you have low data to train your ML models.

Data augmentation or Data manipulation is mostly useful when you have low data.

Training an AI or ML model in a single orientation makes the model learn features only in a single transformation, which will fail when input data is slightly changed. Changes in environmental conditions make the appearance changes and the model may fail to that input, on which it is not trained.

A recommended method is to use multiple transforms of same image, which makes the model robust to the image in any orientation and appearance changes.

Features:

• I can convert a small dataset into a large dataset.
• I can make the dataset invariant to rotation changes, appearance changes, scale variations, etc.

Possible Techniques I can use:

• Rotate
• Flips
• Zoom (or scaling)
• Color Changes
• Blurriness
• Misalignment
• Partial obstructions or occlusions
• Color change
• noise
• bright/dark
• crop
• Gray scale
• shifted