Juliet Fiss1     Aseem Agarwala2     Brian Curless1

1University of Washington    2Adobe Systems, Inc.

Download Video (5 min) [135.7 MB]
Download Paper [4.5 MB]
Download PowerPoint Slides [278 MB]

In this paper, we train a computer to select still frames from video that work well as candid portraits. Because of the subjective nature of this task, we conduct a human subjects study to collect ratings of video frames across multiple videos. Then, we compute a number of features and train a model to predict the average rating of a video frame. We evaluate our model with cross-validation, and show that it is better able to select quality still frames than previous techniques, such as simply omitting frames that contain blinking or motion blur, or selecting only smiles. We also evaluate our technique qualitatively on videos that were not part of our validation set, and were taken outdoors and under different lighting conditions.

Juliet Fiss, Aseem Agarwala, Brian Curless. Candid Portrait Selection From Video. ACM Transactions on Graphics, 2011, 30, 6, (Proc. SIGGRAPH Asia) [BibTex]

Data: includes original validation set videos (and frames) with human ratings [Download - 2.76 GB]
Psychology Study Instructions [Download - 6.9 MB] Photographer Study Instructions [Download - 49 KB]
Psychology Study Setup: includes Java executable and source code for running psychology study [Download - Coming Soon!]
Numerical Results [Download - 33 KB]
Pseudocode: includes pseudocode for our processing pipeline [Download - Coming Soon!]