AutoRoto: Automated Rotoscoping with Refined Deep Masks

The legendary movie Jurassic Park (1993) would have been nearly impossible without Computer Generated Imaginary(CGI). The movie's ground-breaking visual effects was able to smoothly integrate virtual characters with real-life footage. This rapidly increased the use of visual effects in modern movies. Creating high-end visual effects is a complex workflow where each shot is subjected to artistic operations. The key step in this workflow is the process called digital rotoscoping. It is the process of manually outlining and cutting out the foreground characters in a raw video frame via software. This is a time-consuming process and the average number of frames a professional artist can accomplished per day is 15. Due to this, it has become a temporal bottleneck of the VFX pipeline.In order to address the above mentioned issue, an automated rotoscoping approach with instance segmentation is presented in this paper. The novel method is efficient, accurate and will redeem the time of rotoscoping artists. Our model was able to extract foregrounds of multiple instances of classes and to the best of our knowledge our rotoscoping tool is the most generalized to extract foregrounds.