Many Perception Tasks are Highly Redundant Functions of their Input Data

TL;DR

Many perception tasks can be effectively performed using various subspace projections of input data, revealing high redundancy and robustness across different data representations.

Contribution

This paper demonstrates that perception tasks are highly redundant functions of input data, showing effective performance across multiple subspace projections.

Findings

Tasks perform well across different subspaces regardless of data variability.

Redundant information in various subspaces supports task performance.

Different subspaces contain overlapping relevant information.

Abstract

We show that many perception tasks, from visual recognition, semantic segmentation, optical flow, depth estimation to vocalization discrimination, are highly redundant functions of their input data. Images or spectrograms, projected into different subspaces, formed by orthogonal bases in pixel, Fourier or wavelet domains, can be used to solve these tasks remarkably well regardless of whether it is the top subspace where data varies the most, some intermediate subspace with moderate variability--or the bottom subspace where data varies the least. This phenomenon occurs because different subspaces have a large degree of redundant information relevant to the task.