Researchers from the Georgia Institute of Technology have released a dataset of nearly 3,000 algorithmically-augmented videos to improve the ability of systems to detect street signs in difficult conditions. The videos contain 650,000 street signs, each manipulated by the researchers to include simulated weather and camera conditions such as snow, rain, and lens blur. The researchers found that these conditions can reduce the performance of object detection algorithms by 20 to 60 percent.
Image: PxHere