Meta-UDA: Unsupervised Domain Adaptive Thermal Object Detection using Meta-Learning

  • 2021-10-07 02:28:18
  • Vibashan VS, Domenick Poster, Suya You, Shuowen Hu, Vishal M. Patel
  • 23

Abstract

Object detectors trained on large-scale RGB datasets are being extensivelyemployed in real-world applications. However, these RGB-trained models suffer aperformance drop under adverse illumination and lighting conditions. Infrared(IR) cameras are robust under such conditions and can be helpful in real-worldapplications. Though thermal cameras are widely used for military applicationsand increasingly for commercial applications, there is a lack of robustalgorithms to robustly exploit the thermal imagery due to the limitedavailability of labeled thermal data. In this work, we aim to enhance theobject detection performance in the thermal domain by leveraging the labeledvisible domain data in an Unsupervised Domain Adaptation (UDA) setting. Wepropose an algorithm agnostic meta-learning framework to improve existing UDAmethods instead of proposing a new UDA strategy. We achieve this bymeta-learning the initial condition of the detector, which facilitates theadaptation process with fine updates without overfitting or getting stuck atlocal optima. However, meta-learning the initial condition for the detectionscenario is computationally heavy due to long and intractable computationgraphs. Therefore, we propose an online meta-learning paradigm which performsonline updates resulting in a short and tractable computation graph. To thisend, we demonstrate the superiority of our method over many baselines in theUDA setting, producing a state-of-the-art thermal detector for the KAIST andDSIAC datasets.

 

Quick Read (beta)

loading the full paper ...