ECCV 2016 Daily - Friday

On this project work Simone Meyer , PhD student on a collaboration project between ETH and Disney Research, Alexander Sorkine- Hornung, a Senior Research Scientist of Disney Research, and Dr. Markus Gross, a Computer Science professor at ETH Zurich. At ECCV2016, they presented Phase- based Modification Transfer for Video. The project modifies an entire video by editing the first frame, proposing an efficient method to propagate this modification from the first frame across the whole video rather than doing so manually on each frame. The novelty of the project is in the approach they are using. The traditional approach, for example, would be computing optical flow and transferring it. Instead, what they are doing is observing how each pixel changes. The value they observe is the phase which doesn’t need any explicit matching across the frame. The challenge is that it’s a new method which is not well explored leading to unpredictable outcomes. It’s using an old method, but it’s revisited for new applications. It’s also a mathematical formulation. To solve the challenge, the team has experimented by trying out new and different algorithms and combining the information. The only information they have is phase and amplitude and how to combine the two for the application. Sideways, they have the amplitude and the phase. The shift is the motion, but it can implicitly represent the phase of this curve. For the algorithm, as a basis they use steerable pyramid decomposition which gives the phase and amplitude value. Thereon, it’s its own algorithm. The practical application is that it can be used for any frame editing. It’s really suitable for high frame rate video because it can handle only tiny motion. It needs frames with small motions in between. It’s also suited for high resolution data because it works so fast. Presentations 21 ECCV Daily: Fr iday “ The practical application is that it can be used for any frame editing ” Phase-based Modification Transfer for Video