Systems and methods for use in training a convolutionalneural network (CNN) for image and video transformations. TheCNN is trained by adding noise to training data set images,transforming both the noisy image and the source image, and thendetermining the difference between the transformed noisy imageand the transformed source image. The CNN is further trained byusing an object classifier network and noting the nodeactivation levels within that classifier network whentransformed images (from the CNN) are classified. Byiteratively adjusting the CNN to minimize a combined lossfunction that includes the differences between the nodeactivation levels for the transformed references images and whentransformed source are classified and the differences betweenthe transformed noisy image and the transformed source image,the artistic style being transferred is maintained in thetransformed images.
展开▼