Corresponding Author: Nisha Pal
Computer vision has generally been guided through deep learning, primarily in terms of thinking about devices in snapshots. Convolutional neural community-based absolute models have completely changed the panorama. Models select features in their own unique way, saving countless hours of manual work. Object identification involves more than just identifying equipment in an image; It also contains a clue to their special locations. This has become more important for medical imaging or driverless cars, even to more intelligent retail systems. However, there is a problem. Deep learning models require vast, diverse, and intensive datasets. The process of gathering and labeling images is costly and time-consuming. This is when data augmentation is useful. To expand their raw datasets, people employ a variety of techniques such image flipping, rotation, cropping, and brightness adjustment. It is really beneficial, but those adjustments are really easy. For instance, they don't deal with situations when one object is in front of another, complicated backgrounds, dim lighting, or even instances where some categories are hardly represented in the data. As a result, models continue to struggle with the chaos that occurs outside of the laboratory. The concept was used to develop a more sophisticated pipeline for data augmentation. This means that occlusions should be simulated, context should be incorporated, and the changes that are most important for the task should be added. As a result, the model won't become stuck on training data and will learn better patterns.