Research Commons
      • Browse 
        • Communities & Collections
        • Titles
        • Authors
        • By Issue Date
        • Subjects
        • Types
        • Series
      • Help 
        • About
        • Collection Policy
        • OA Mandate Guidelines
        • Guidelines FAQ
        • Contact Us
      • My Account 
        • Sign In
        • Register
      View Item 
      •   Research Commons
      • University of Waikato Research
      • Computing and Mathematical Sciences
      • Computing and Mathematical Sciences Papers
      • View Item
      •   Research Commons
      • University of Waikato Research
      • Computing and Mathematical Sciences
      • Computing and Mathematical Sciences Papers
      • View Item
      JavaScript is disabled for your browser. Some features of this site may not work without it.

      Domain specific transfer learning using image mixing and stochastic image selection

      Coup, Sheldon; Vetrova, Varara; Frank, Eibe; Tappenden, Rachael
      Thumbnail
      Files
      FGVC6 accepted paper.pdf
      Accepted version, 136.0Kb
      Link
       drive.google.com
      Citation
      Export citation
      Coup, S., Vetrova, V., Frank, E., & Tappenden, R. (2019). Domain specific transfer learning using image mixing and stochastic image selection. Presented at the The Sixth Workshop on Fine-Grained Visual Categorization (FGVC6), Computer Vision and Pattern Recognition Conference (EVPR 2019), Long Beach, CA.
      Permanent Research Commons link: https://hdl.handle.net/10289/13609
      Abstract
      Can a gradual transition from the source to the target dataset improve knowledge transfer when fine-tuning a convolutional neural network to a new domain? Can we use training examples from general image datasets to improve classification on fine-grained datasets? We present two image similarity metrics and two methods for progressively transitioning from the source dataset to the target dataset when fine-tuning to a new domain. Preliminary results, using the Flowers 102 dataset, show that the first proposed method, stochastic domain subset training, gives an improvement in classification accuracy compared to standard fine-tuning, for one of the two similarity metrics. However, the second method, continuous domain subset training, results in a reduction in classification performance.
      Date
      2019
      Type
      Conference Contribution
      Collections
      • Computing and Mathematical Sciences Papers [1454]
      Show full item record  

      Usage

      Downloads, last 12 months
      18
       
       

      Usage Statistics

      For this itemFor all of Research Commons

      The University of Waikato - Te Whare Wānanga o WaikatoFeedback and RequestsCopyright and Legal Statement