Supplementary document: Transient Attributes for High-Level Understanding and Editing of Outdoor Scenes

Our Transient Attribute Database: selection of representative images (Section 1.1), screen captures of our crowdsourced annotation tasks (Section 1.2), comparison of our aggregated labels to meteorological data (Section 1.3), and correlation between attributes (Section 1.4). Our attribute recognition method: description of the image features and encoding methods used (Section 2). In addition, our project website contains: • the list of our 40 transient attributes along with their definition and positive/negative examples • all 8571 images from our database with their attribute labels • our trained attribute regressors (with Fisher Vector encoding and spatial pyramids), along with train/test splits • results of our appearance transfer method on 60 test cases for different attribute manipulation queries, along with the match/target images used, and comparison to three other methods