Unsupervised Deep Domain Adaption for Object Detection

Aug 15, 2019
PDF
Foggy Cityscape Dataset
  • Analyzed negative transfer (around 20% drop in mAP from baseline) and catastrophic forgetting of the existing imageto-image domain adaptation approaches on face-detection datasets.
  • Studied the use of local features, and temporal information from trackers to generalize unsupervised domain adaptation approaches on datasets like SCUT and Widerface.
  • Advisor: Eric Granger
Deep Learning Computer Vision
Benedict Florance Arockiaraj
Benedict Florance Arockiaraj
ML Engineer

My research interests are at the juncture of deep learning and computer vision.