Photo Organization & Manipulation

Thursday, 28 July, 3:45 pm - 5:15 pm, Anaheim Convention Center, Ballroom C
Session Chair: Ping Tan, Simon Fraser University

Automatic Triage for a Photo Series

Introducing a method for ranking photos taken in the same scene. After collecting a dataset of photo series from unedited personal collections, the authors asked people to rank them, then applied new machine-learning approaches for modeling human preference.

Huiwen Chang
Princeton University

Fisher Yu
Princeton University

Jue Wang
Adobe Research

Douglas Ashley
Princeton University

Adam Finkelstein
Princeton University

Automatic Photo Adjustment Using Deep Neural Networks

This work formulates photo adjustment as a highly nonlinear regression problem, which can be effectively solved by deep neural networks fed with proposed contextual feature descriptors. The paper demonstrates that deep neural networks using these descriptors can successfully capture sophisticated photographic styles.

Zhicheng Yan
University of Illinois at Urbana-Champaign

Hao Zhang
Carnegie Mellon University

Baoyuan Wang
Microsoft Research

Sylvain Paris
Adobe Systems Incorporated

Yizhou Yu
University of Hong Kong, University of Illinois at Urbana-Champaign

EyeOpener: Editing Eyes in the Wild

This paper presents a framework for automatically replacing “bad” eyes in photographs. The pipeline includes automatic reference selection based on crowdsourced evaluation statistics, 3D face estimation, warping, harmonization, and compositing.

Zhixin Shu
Stony Brook University

Eli Shechtman
Adobe Research

Dimitris Samaras
Stony Brook University

Sunil Hadap
Adobe Research

Sky is Not the Limit: Semantic-Aware Sky Replacement

Skies are common backgrounds in many photos, but are often uninteresting. Artists correct this using complicated workflows that are beyond causal users. This work proposes a system that can automatically replace sky with semantic guidance and produce a set of diverse, realistic, and visually pleasing results.

Yi-Hsuan Tsai
University of California, Merced

Xiaohui Shen
Adobe Research

Zhe Lin
Adobe Research

Kalyan Sunkavalli
Adobe Research

Ming-Hsuan Yang
University of California, Merced