Towards Large-Scale Video Video Object Mining

TL;DR

This paper introduces a method for large-scale object mining in unlabeled automotive videos using object tracking, resulting in a new dataset and preliminary detector adaptation results.

Contribution

It presents a novel approach combining object tracking with automated category discovery and detector learning in large-scale unlabeled video data.

Findings

Created a dataset of over 360,000 object tracks

Demonstrated automated discovery of novel object categories

Showed initial success in detector adaptation using mined tracks

Abstract

We propose to leverage a generic object tracker in order to perform object mining in large-scale unlabeled videos, captured in a realistic automotive setting. We present a dataset of more than 360'000 automatically mined object tracks from 10+ hours of video data (560'000 frames) and propose a method for automated novel category discovery and detector learning. In addition, we show preliminary results on using the mined tracks for object detector adaptation.