Regularization in Computer Vision – Synthetic Image Generation
This chapter will focus on the techniques and methods used to generate synthetic images for data augmentation. Having diverse data is often one of the most efficient ways to regularize computer vision models. Many approaches allow us to generate synthetic images; from simple tricks such as image flipping to new image creation using generative models. Several techniques will be explored in this chapter, including the following:
- Image augmentation with Albumentations
- Creating synthetic images for object detection – training an object detection model with only synthetic data
- Real-time style transfer – training a model for real-time style transfer based on Stable Diffusion, a powerful generative model
