View Based Methods can achieve Bayes-Optimal 3D Recognition

TL;DR

This paper demonstrates that view-based 2D similarity methods can match the performance of 3D model-based recognition systems and do not require more training data, challenging assumptions about the necessity of 3D models.

Contribution

It proves that view-based recognition systems can achieve Bayes-optimal performance without using 3D models or additional training views.

Findings

View-based methods can match 3D model-based recognition performance.

Training data requirements are similar for both view-based and 3D systems.

Performance alone cannot distinguish between view-based and 3D model-based recognition.

Abstract

This paper proves that visual object recognition systems using only 2D Euclidean similarity measurements to compare object views against previously seen views can achieve the same recognition performance as observers having access to all coordinate information and able of using arbitrary 3D models internally. Furthermore, it demonstrates that such systems do not require more training views than Bayes-optimal 3D model-based systems. For building computer vision systems, these results imply that using view-based or appearance-based techniques with carefully constructed combination of evidence mechanisms may not be at a disadvantage relative to 3D model-based systems. For computational approaches to human vision, they show that it is impossible to distinguish view-based and 3D model-based techniques for 3D object recognition solely by comparing the performance achievable by human and 3D model-based systems.}