Localizing Occluders with Compositional Convolutional Networks

TL;DR

This paper improves compositional convolutional networks' ability to localize occluders in images by modeling feature activations with von-Mises-Fisher distributions, enhancing occlusion handling without requiring occluded training data.

Contribution

The paper introduces a novel modeling approach using von-Mises-Fisher distributions for compositional networks, enabling end-to-end training and better occluder localization.

Findings

Enhanced occluder localization accuracy

Improved classification of partially occluded objects

Effective end-to-end training of the model

Abstract

Compositional convolutional networks are generative compositional models of neural network features, that achieve state of the art results when classifying partially occluded objects, even when they have not been exposed to occluded objects during training. In this work, we study the performance of CompositionalNets at localizing occluders in images. We show that the original model is not able to localize occluders well. We propose to overcome this limitation by modeling the feature activations as a mixture of von-Mises-Fisher distributions, which also allows for an end-to-end training of CompositionalNets. Our experimental results demonstrate that the proposed extensions increase the model's performance at localizing occluders as well as at classifying partially occluded objects.