ICTree: Automatic Perceptual Metrics for Tree Models

Many algorithms for synthetic tree generation exist, but the visual quality of the generated models is unknown. This problem is usually solved by performing limited user studies or by side-by-side comparison. We introduce an automated system for quality assessment of the tree model based on their perception. We conducted a user study in which over one million pairs of images were compared to collect subjective perceptual scores of generated trees. The perceptual score was used to train two neural-network-based predictors. A view independent ICTreeF uses the tree models geometric features that are easy to extract from any model. The second is ICTreeI that estimates the perceived visual quality of a tree from its image. Moreover, to provide an insight into the problem, we deduce intrinsic attributes and evaluate which features make trees look like real trees. In particular, we show that branching angles, length of branches, and widths are critical for perceived realism.