Comprehensive analysis of shot boundary detection methods

This benchmark is designed to evaluate different algorithms of the Shot Boundary Detection task. The benchmark provides an opportunity to test your own algorithm for detection of cuts, dissolves, wipes and fades scene transition.

Key features of the Benchmark

• Extensive and diverse datasets for shot-boundary detection testing:
  • 120+ videos from 2 datasets: RAI, MSU CC
  • 1000+ minutes with 1700+ shot transitions
  • Cut, dissolve and fade transitions, complex scenes with high spatial and temporal complexity
• Beautiful and easy-interpreting visualizations of the most common quality metrics for each category of scene transitions:
  • F1 score, Precision and Recall
  • Speed performance depending on the resolution of the input video stream

To participate in our benchmark, please follow the instructions in the How to submit section.

We appreciate new ideas. Please, write us an e-mail to sbd-benchmark@videoprocessing.ai

Leaderboard

The table below shows a comparison of SBD algorithms by F1 score and speed.

Performance Chart

Definitions

• Shot — the continuous footage or sequence between two edits or cuts
  • Cut Transitions — a sudden abrupt transition from one shot to another
    • In this case, we mark the first frame of a new shot as a shot with a cut transition
• Gradual Transitions — two shots are combined using chromatic, spatial or spatial-chromatic effects which gradually replace one shot with another
  • In this case we mark a continuous series of frames as a gradual transition. Algorithms must detect exactly one frame as a gradual transition from this series
• Fade-out and fade-in Transitions are used to describe a transition to and from a blank image

How to submit

Send us an email to sbd-benchmark@videoprocessing.ai with the following:

• File with .exe extension with your algorithm and instructions to run it from the console. When launched from console we expect your file to have the following options:
  • -i - path to input video
  • -o - path to the output of your algorithm
  • -t - threshold, if required in your algorithm
• If you give us permission to publish the results of your algorithm.

If you would like to participate read the full submission rules here.

Contacts

For questions and propositions, please contact us: sbd-benchmark@videoprocessing.ai

Learn more