Back to Basics: Unsupervised Learning of Optical Flow via Brightness Constancy and Motion Smoothness

TL;DR

This paper introduces an unsupervised convolutional neural network approach for optical flow estimation that relies on brightness constancy and motion smoothness, eliminating the need for labeled datasets.

Contribution

It presents a novel unsupervised training method for optical flow prediction using photometric and smoothness constraints, outperforming supervised methods on KITTI.

Findings

Unsupervised training achieves better results than supervised on KITTI.

The approach reduces reliance on expensive labeled datasets.

Convnet trained with the proposed loss functions effectively predicts optical flow.

Abstract

Recently, convolutional networks (convnets) have proven useful for predicting optical flow. Much of this success is predicated on the availability of large datasets that require expensive and involved data acquisition and laborious la- beling. To bypass these challenges, we propose an unsuper- vised approach (i.e., without leveraging groundtruth flow) to train a convnet end-to-end for predicting optical flow be- tween two images. We use a loss function that combines a data term that measures photometric constancy over time with a spatial term that models the expected variation of flow across the image. Together these losses form a proxy measure for losses based on the groundtruth flow. Empiri- cally, we show that a strong convnet baseline trained with the proposed unsupervised approach outperforms the same network trained with supervision on the KITTI dataset.