Scalable Full Flow with Learned Binary Descriptors

TL;DR

This paper introduces a scalable method for large displacement optical flow using learned binary descriptors and a novel min-projection technique, enabling efficient high-resolution processing and accurate learning.

Contribution

It presents a new hybrid learning scheme for binary CNN descriptors and a min-projection method that reduces memory complexity, allowing high-resolution optical flow estimation.

Findings

Efficient binary descriptors enable on-the-fly cost evaluation.

The min-projection reduces memory from quadratic to linear.

The method achieves competitive results on benchmarks.

Abstract

We propose a method for large displacement optical flow in which local matching costs are learned by a convolutional neural network (CNN) and a smoothness prior is imposed by a conditional random field (CRF). We tackle the computation- and memory-intensive operations on the 4D cost volume by a min-projection which reduces memory complexity from quadratic to linear and binary descriptors for efficient matching. This enables evaluation of the cost on the fly and allows to perform learning and CRF inference on high resolution images without ever storing the 4D cost volume. To address the problem of learning binary descriptors we propose a new hybrid learning scheme. In contrast to current state of the art approaches for learning binary CNNs we can compute the exact non-zero gradient within our model. We compare several methods for training binary descriptors and show results on public available benchmarks.