ECCV 2016 Daily - Wednesday

To solve this, they used Eigen Textures , a previously existing technique that re-factors a set of texture. In other words, it captures the variation space of a set of textures with a PCA. They employed this kind of technique with some enhancements. They also compute warps to realign textures with small geometric variations within the sequence. They have factored in some portions of the variation, but they are looking ahead and think more things can be done. For example, while they have already re-factored micro- geometries and recapture variations once the geometries align, they could more explicitly model light variations. They could try to capture one albedo irrespective of where the light source is. They could also separate the lighting component. In terms of the practical application, their model can generate virtual characters. With several sequences of the same person, it can generate full virtual characters by re- rendering the representation to generate new viewpoints. For example, in recorded sequences in which the character jumps, runs, and moves, it can transition between these different sequences and animations and generate new animations. It could provide a full character model and do renderings of this character model by interpolating between the acquired sequences. Their technique provides an efficient way to do this because, with Eigen texture representation, it can easily interpolate between sequences by interpolating the coefficients of the Eigen representation between the sequences. The other application of this work is for texture completion. Let’s say you have a sequence of an actor whose performance you are capturing, and some part of his body is being included for that particular view. Since there is a texture space or PCA space that captures the variability, it can reproject that particular frame in the PCA space and generate the missing part of the texture by extrapolating it from the texture space. Presentations 9 ECCV Daily: Wednesday