Partial Duplicate Video Retrieval based on Deep Metric Learning

View Paper
  • Extracted video feature through VGG-16 backbone network and applied Deep Metric Learning (DML) for video similarity calculation.
  • Proposed a video segmentation method based on the above extracted video feature using a DNN model, which can effectively segment different scenes in a long video.
  • Developed a multi-thread downloader for efficiently downloading the large CC_WEB_VIDEO dataset and built a segmentation dataset from Bilibili and Youtube, which contains 36,600 samples.
PDVR DML Pipeline VGG structure DNN Structure