Rotoscope Automation with Deep Learning

Oscar Estrada Torrejon, Nicholas Peretti, Ricardo Figueroa

We present a deep learning-based algorithm that can automatically rotoscope people in a given scene, without any user input. Current approaches to image matting require a significant amount of human input, irrespective of whether it is manually rotoscoped or through a chroma key. This study shows that this algorithm can perform as well as and even surpass the rotoscoping capabilities of the Adobe After Effects’ RotoBrush tool, in a variety of scenes comprising different lighting conditions, movements, and subjects. This makes it suitable for integration within a visual effects (VFX) pipeline.

Print ISSN
Electronic ISSN
2160-2492
Published
2020-03
Content type
Original Research
Keywords
Compositing, deep learning, image matting, instance segmentation
DOI
10.5594/JMI.2019.2959967