augmenting large datasets to have more edge case data for training [D]

A user proposes augmenting large datasets to include more edge case data for training models. The idea involves adapting existing labeled datasets, which primarily consist of sunny daytime footage, to simulate challenging conditions like night, fog, rain, or glare. This augmentation would use physics-based effects and constrained generative models to transform clear daytime HD driving footage into scenarios resembling a cheap dashcam at night in the rain with glare and heavy compression, while keeping labels intact.