Fast and robust visual tracking with hard balanced focal loss and guided domain adaption
作者：
Highlights：
• A guided domain adaptation module is proposed, which adapt the offline trained general domain to the tracking target domain.
• A hard balanced focal loss function is developed to alleviate the positive–negative and the easy-hard sample imbalance.
• We select the Fully Connected Siamese network as the baseline and our tracker obtain high progress on six popular benchmarks.
摘要
•A guided domain adaptation module is proposed, which adapt the offline trained general domain to the tracking target domain.•A hard balanced focal loss function is developed to alleviate the positive–negative and the easy-hard sample imbalance.•We select the Fully Connected Siamese network as the baseline and our tracker obtain high progress on six popular benchmarks.
论文关键词：Visual tracking,Siamese network,Domain adaptation,Hard balanced focal loss
论文评审过程：Received 9 April 2020, Accepted 3 May 2020, Available online 14 May 2020, Version of Record 25 June 2020.
论文官网地址：https://doi.org/10.1016/j.imavis.2020.103929
