Unaligned Image-to-Sequence Transformation with Loop Consistency

TL;DR

This paper introduces a method for modeling sequential visual phenomena without requiring aligned data, extending cycle consistency to loop consistency to generate sequences from unpaired images, demonstrated on datasets like seasons and aging faces.

Contribution

It proposes a novel loop consistency framework for unpaired image-to-sequence transformation, enabling sequence modeling without aligned ground-truth data.

Findings

Achieved competitive results on seasonal and aging datasets.

Extended cycle consistency to loop consistency for long sequence modeling.

Demonstrated effectiveness without aligned sequence data.

Abstract

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.