Unaligned Image-to-Sequence Transformation with Loop Consistency
This addresses the challenge of generating sequences without aligned data, which is incremental as it extends unpaired image-to-image techniques to sequences.
The paper tackled the problem of modeling sequential visual phenomena by generating sequences from unaligned images across time steps, showing competitive results on datasets like Earth's seasons and human face aging.
We tackle the problem of modeling sequential visual phenomena. Given examples of a phenomena that can be divided into discrete time steps, we aim to take an input from any such time and realize this input at all other time steps in the sequence. Furthermore, we aim to do this without ground-truth aligned sequences -- avoiding the difficulties needed for gathering aligned data. This generalizes the unpaired image-to-image problem from generating pairs to generating sequences. We extend cycle consistency to loop consistency and alleviate difficulties associated with learning in the resulting long chains of computation. We show competitive results compared to existing image-to-image techniques when modeling several different data sets including the Earth's seasons and aging of human faces.