Unsupervised Moving Object Detection via Contextual Information Separation

TL;DR

This paper introduces an unsupervised adversarial model that detects moving objects by separating contextual information, outperforming supervised methods without needing annotated data.

Contribution

It presents a novel unsupervised adversarial framework for moving object detection that eliminates the need for regularization and hyper-parameter tuning.

Findings

Outperforms several supervised methods on moving object detection tasks.

Requires no annotated data for training.

Generalizes classical segmentation approaches without explicit regularization.

Abstract

We propose an adversarial contextual model for detecting moving objects in images. A deep neural network is trained to predict the optical flow in a region using information from everywhere else but that region (context), while another network attempts to make such context as uninformative as possible. The result is a model where hypotheses naturally compete with no need for explicit regularization or hyper-parameter tuning. Although our method requires no supervision whatsoever, it outperforms several methods that are pre-trained on large annotated datasets. Our model can be thought of as a generalization of classical variational generative region-based segmentation, but in a way that avoids explicit regularization or solution of partial differential equations at run-time.