A simpler path to better computer vision
Research finds using a large collection of simple, un-curated synthetic image generation programs to pretrain a computer vision model for image classification yields greater accuracy than employing other pretraining methods that are more costly and time consuming, and less scalable.
Manel Baradad Antonio Torralba Rogerio Feris Tongzhou Wang Phillip Isola Artificial Intelligence Laboratory
Source: mit.edu