TMK

Adapted the kernel descriptor framework of Bo (paper) to sequences of frames. Proposed a query expansion (QE) technique that automatically aligns the videos deemed relevant for the query.

Added to the benchmark by MSU G&M Lab

Links: GitHub, paper

VideoIndexer

Detect scene changes, split the video on scenes. Align scenes, than align frames respectively.

Added to the benchmark by MSU G&M Lab

Links: paper

Time shift metric in VQMT tool

Use PSNR to detect relevant frames.

Added to the benchmark by MSU G&M Lab

Links: project

Time shift metric in VQMT3D tool

Use motion vectors and RANSAC to measure time shift between frames.

Added to the benchmark by MSU G&M Lab

Links: project

ViSiL

Use RMAC descriptors to estimate frame-to-frame and video-to-video similarity.

Added to the benchmark by MSU G&M Lab

This method was modified by MSU to suit the benchmark suite tasks: we measure only frame-to-frame similarity, then we make synchronization map by taking maximum values in the resulting cost matrix.

Links: GitHub, paper

ViSiL_SCD

Use ViSiL architecture to compute frames features. Detect scene changes by the features and split videos on scenes. Match the scenes by video-to-video similarity and then make synchronization map by taking maximum values in the frame-to-frame similarity matrix.

Added to the benchmark by MSU G&M Lab