Abstract

Master photographers such as Ansel Adams have demonstrated that retouching pictures is critical to obtaining a stunning rendition of a picture. In traditional film photography, this involves tools such as a piece of cardboard to dodge and burn the print as well as a careful choice of chemicals and paper. This is a long and painstaking process that requires a great deal of know-how. With digital photography, we have now access to powerful options such as instant undo, unlimited layering, and sophisticated filters. However, despite these improvements, achieving a compelling look remains a challenging task. In some sense, we have made the piece of cardboard more powerful, which is great and enables a wider range of effects, but the retouching process as a whole remains mostly unassisted. In this talk, I will present my research on photographic style transfer and machine-learning-based retouching that focuses on understanding the notion of photographic look and helping users achieve a specific visual style. This work is in collaboration with Soonmin Bae, Vladimir Bychkovsky, and Frédo Durand at MIT , and Eric Chan at Adobe.