MSU Video Alignment and Retrieval Benchmark Suite

Explore the best algorithms in different video alignment tasks

Powered by
G&M Lab head: Dr. Dmitriy Vatolin
Measurements, analysis: 
Mikhail Dremin,
Konstantin Kozhemyakov

What’s new

  • 22.10.2021 Public beta-version Release
  • 06.10.2021 Alpha-version Release

Key features of the Benchmark Suite

  • The most diverse dataset for alignment of near-duplicate videos:
    • 560 test pairs in each Benchmark with a total duration of ~2 million frames
    • Combinations of 13 frequent distortions obtained due to human/machine video editing and processing
  • Test your method on our Benchmarks:
    • Local time distortions: occur due to video processing (transmitting, compressing with low bitrate etc.)
    • Global time distortions: occur due to video editing
    • Mixed version for versatility testing on both distortion types
  • Find the best method for your near-duplicate video alignment requirements:

To submit your method, please, follow several simple steps in the Algorithm Submission section

We appreciate new ideas. Please, write us an e-mail to


You can choose the preset on which the algorithms were tested and sort charts by the algorithm.

NOTE: The same number of test pair on different charts does not mean the same test pair because of sorting the results by algorithm.

Local Time Distortions


Name Accuracy
0 frames error
3 frames error
10 frames error


Highlight the plot region where you want to zoom in

Global Time Distortions


Name F1-score Precision Recall


Highlight the plot region where you want to zoom in

Mixed Time Distortions


Name F1-score Precision Recall


Highlight the plot region where you want to zoom in


Video sequences selection

Content diversity in test sequences is necessary for running algorithms in near-realistic conditions. To provide this diversity the following steps have been undertaken.

We identified two families of cases: processes during which temporary distortions could occur, i.e. video editing, streaming, capturing from different rigidly-connected cameras etc., and content in test videos which is difficult to align, i.e. talking head, recurring events (such as football matches, formula-1 racing), camera on a tripod (small amount of movement) etc. Then, we collected ~2000 videos from Vimeo from each taxonomic category to ensure unbiased video selection and sufficient coverage in terms of semantics and context presented in the video.

Finally, SI/TI features were calculated for each video and for each class we mapped videos to use cases mentioned above.

We chose videos that sufficiently cover SI/TI space and all the identified use cases and got 56 source videos.


Our dataset is constantly updated. Now we have 56 source video sequences with total duration of 195,411 frames. Resolution of all video sequences is 1920×1080. FPS ranges from 16 to 60.

Benchmarks and presets

We decided to divide the problem of video alignment into three logical parts: local time distortions (occurs due to video processing), global time distortions (occurs due to video editing) and mixed time distortions (for versatility testing on both distortion types). Hence, our benchmark suite contains three benchmarks.

Benchmark Time distortions Metric
Local time distortions Drop frames Accuracy(%) of correctly localized time-shifts of initial frames depending on maximum accepted error in frames
Duplicate frames
Freeze frames
Time shift
Global time distortions Fragment insertion F1-score
Drop fragment
Cut the video
Mixed time distortions All time distortions

Metrics are calculated as follows:

Each benchmark consists of four presets for a more precise understanding of the pros and cons of algorithms.

Preset Used distortions
Light Time distortions
Medium geometric Time distortions
Geometric distortions
Medium color Time distortions
Color distortions
Hard All distortions

Distortions and distortion distribution

Distortion type Distortion Description
Time distortions Drop frames First, last or random frame number < fps, drop/duplicate 1, 2 or random up to 10 frames
Duplicate/Freeze frames
Time shift Crop up to 11 frames from the start
Fragment insertion Insert fragment from another video
Drop fragment Drop random fragment
Cut the video Cut up to 3/4 of the entire video
Color distortions Add noise Add gaussian noise, sigma 3
Add blur Add gaussian blur, sigma 1.5
Random brightness&contrast Random brignthess decrease from -10% up to -25%,
contrast increase from 5% up to 30%
Geometric distortions Add logo Logo or news title. Covers max 300px from bottom/top/left/right
Add subtitles English language. Covers max 250px from bottom or top
Crop Crop 2/4/8 px from right and bottom
Scale Scale up to 110%
Add black bars Crop to a ratio of 4:3 or 21:9. Then return to 16:9 by adding black bars

Submit your algorithm

1. Download input data Download the necessary folders for the benchmark you want to take part in here.
Some important notes:
    A. Each benchmark has train and validation sets
    B. There are 3 folders for each benchmark, containing 4 zip-folders for each preset

2. Apply your algorithm Align videos with your algorithm
You can also send us the code of your method or the executable file and we will run it ourselves.
We expect a specific output of your algorithm. The description and some examples can be found in
downloadable dataset folder. Depending on benchmark, we calculate Accuracy, F1-score,
Precision and Recall.

3. Send us result Send us an email to with the following information:
    A. Name of your method that will be specified in our benchmark
    B. Link to the cloud drive (Google Drive, OneDrive, Dropbox, etc.), containing output maps
    C. (Optional) Any additional information about the method:
      1. Full name of your model
      2. The parameter set that was used
      3. A link to the code of your model, if it is available
      4. A link to the paper about your model
      5. Execution time of your algorithm and information about used GPU and CPU
      6. Any other additional information


For questions and propositions, please contact us:

MSU Video Quality Measurement Tool


    The tool for performing video/image quality analyses using reference or no-reference metrics

Widest Range of Metrics & Formats

  • Modern & Classical Metrics SSIM, MS-SSIM, PSNR, VMAF and 10+ more
  • Non-reference analysis & video characteristics
    Blurring, Blocking, Noise, Scene change detection, NIQE and more

Fastest Video Quality Measurement

  • GPU support
    Up to 11.7x faster calculation of metrics with GPU
  • Real-time measure
  • Unlimited file size

  • Main MSU VQMT page on

22 Oct 2021
See Also
Real-World Stereo Color and Sharpness Mismatch Dataset
Download new real-world video dataset of stereo color and sharpness mismatches
Super-Resolution Quality Metrics Benchmark
Discover 66 Super-Resolution Quality Metrics and choose the most appropriate for your videos
Learning-Based Image Compression Benchmark
The First extensive comparison of Learned Image Compression algorithms
Video Saliency Prediction Benchmark
Explore the best video saliency prediction (VSP) algorithms
Super-Resolution for Video Compression Benchmark
Learn about the best SR methods for compressed videos and choose the best model to use with your codec
Metrics Robustness Benchmark
Check your image or video quality metric for robustness to adversarial attacks
Site structure