Self-supervised AutoFlow

TL;DR

Self-supervised AutoFlow is a novel method that automatically searches for optimal training configurations for optical flow learning using self-supervised losses, eliminating the need for ground truth labels and achieving competitive results on multiple datasets.

Contribution

It introduces a self-supervised search metric for AutoFlow, enabling training set optimization without ground truth labels and improving performance on real-world videos.

Findings

Performs on par with AutoFlow on Sintel and KITTI datasets.

Outperforms AutoFlow on the DAVIS dataset.

Achieves competitive results in semi-supervised settings.

Abstract

Recently, AutoFlow has shown promising results on learning a training set for optical flow, but requires ground truth labels in the target domain to compute its search metric. Observing a strong correlation between the ground truth search metric and self-supervised losses, we introduce self-supervised AutoFlow to handle real-world videos without ground truth labels. Using self-supervised loss as the search metric, our self-supervised AutoFlow performs on par with AutoFlow on Sintel and KITTI where ground truth is available, and performs better on the real-world DAVIS dataset. We further explore using self-supervised AutoFlow in the (semi-)supervised setting and obtain competitive results against the state of the art.