MSU Deinterlacer Benchmark — selecting the best deinterlacing filter
The most comprehensive comparison of deinterlacing methods
- 26.11.2020 Beta-version Release
- 22.12.2020 Added Adobe Premiere Built-In, VS EEDI3, VS TDeintMod, MC Deinterlacer, EL MAB. Tuned Kernel Deinterlacer
Key features of the Benchmark
- For Deinterlacing methods’ users
- Choose deinterlacing method that is the best for your speed and quality requirements
- Discover the newest deinterlacing methods’ achievements
- For Researchers and Developers
- Quickly get comprehensive comparison results for your paper with our tables, visual comparison tools and performance plots
- Check the performance of your deinterlacing method on the complex cases
To submit deinterlacing method, please, follow 3 simple steps in the the Deinterlacer Submission section
We appreciate new ideas. Please, write us an e-mail to firstname.lastname@example.org
The table below shows a comparison of deinterlacers by PSNR, SSIM metrics and by speed.
Click on the labels to sort the table
|Rank||Name||PSNR||SSIM||FPS on CPU|
|6.0||Real-Time Deep Deinterlacer||39.203||0.976||0.27|
|8.5||Weston 3-Field Deinterlacer||38.680||0.969||36.75|
|9.0||Kernel Deinterlacer (optimal parameters)||38.103||0.970||37.91|
|9.0||Elemental Live Low Latency Interpolation||38.056||0.972||Hardware Real-Time|
|12.0||Elemental Live Motion Adaptive Interpolation||37.063||0.964||Hardware Real-Time|
|14.5||Studio Coast Pty vMix||36.990||0.950||Hardware Real-Time|
|14.5||Adobe Premiere Pro Built-In||36.092||0.958||43.82|
|16.0||Motion and Area Pixel Deinterlacer||35.415||0.950||2.15|
|19.5||Elemental Live Motion Adaptive Blend||29.744||0.868||Hardware Real-Time|
|20.5||Motion Compensation Deinterlacer||27.899||0.804||1.45|
Full FrameRate Leaderboard
|Rank||Name||PSNR||SSIM||FPS on CPU|
|4.5||Real-Time Deep Deinterlacer||39.450||0.977||0.27|
|7.0||Weston 3-Field Deinterlacer||38.726||0.969||36.75|
|7.5||Elemental Live Low Latency Interpolation||37.908||0.972||Hardware Real-Time|
|10.0||Elemental Live Motion Adaptive Interpolation||36.953||0.964||Hardware Real-Time|
|11.0||Studio Coast Pty vMix||32.942||0.931||Hardware Real-Time|
|12.0||Elemental Live Motion Adaptive Blend||29.747||0.868||Hardware Real-Time|
In this section you can see a frame, a crop from this frame, and also MSU VQMT PSNR Visualization of this crop.
Drag a red rectangle in the area which you want to crop, by default it is in the area with the worst PSNR
The area to compare on: Deinterlacer 1: Deinterlacer 2:
In this row you can see VQMT PSNR Visualization
Highlight the plot region where you want to zoom in
FPS is calculated on Intel-Core i7 10700K CPU
The following plot shows difference between every method and Bob, because Bob is considered as the least complicated deinterlacing method
Sequence №: Metric:
In this section you can see PSNR between the output of chosen deinterlacer and the others
Our dataset is constantly updated. Now we have 40 video sequences. Each sequence's length is 1 second. Resolution of all video sequences is 1920x1080. FPS varies from 24 to 60. TFF interlacing was used to get interlaced data from GT.
Click on this text to read how exactly our dataset was composed.
- Initially we had about 30 videos from Vimeo 90k dataset, total length of which was about 1 hour. These videos included sports, panorama, news, landscapes, parts of movies, ads and other types of content.
- We interlaced all these videos, and then measured PSNR between interlaced video and odd frames of gt.
- From each video, we wanted to get 1 or 2 sequences, each 1 second long.
- The sequences with the smallest mean PSNR were considered as the most “damaged”. Our hypothesis was that these sequences will be the hardest for deinterlacers.
- Also we counted mean PSNR over all videos, and the sequences with the closest mean PSNR to mean-of-all PSNR was considered as the most “average”.
- The sequences with the highest mean PSNR were considered as the most “undamaged”. These sequences were often a static shot with no moving objects.
- We took 15 “damaged”, 20 “average”, and 5 “undamaged” sequences and put them together in one video, but separated by 10 black frames.
- We don’t count PSNR on black frames. They were made for deinterlacers that use motion estimation (ME). Because of the black frames, ME-deinterlacer detects zero motion between two sequences and, therefore, doesn’t consider motion from the previous sequence while processing the current one. Also, we ignore the first 4 frames of each sequence while computing the overall mean. Again, we suppose that on the first 4 frames ME-deinterlacers are collecting information about motion and don't show their best.
We compare RGB frames via 2 metrics - PSNR and SSIM. PSNR and SSIM are measured over the Y-component.
For each video sequence, we take the average PSNR and SSIM over all frames. We decided to choose these metrics because they proved themselves to be among the best metrics to show quality loss.
We decided to measure these metrics over Y component because YUV is the most popular type of colorspace nowadays, but there are still a lot of versions of YUV (e.g. yuv444p, yuv420p, yuv420p12le). In these versions U and V components are different, that’s why we measure only Y component. Also, there are a lot of other color spaces, which use Y component (e.g. YCbCr, YPbPr, UYVY, ...). Finally, we can easily compute Y component from other color spaces, such as RGB or gray.
Here is the plot of the PSNR difference between PSNR-Y and PSNR-RGB. As we can see here, the difference is rather negligible.
Validation of deinterlacers' outputs
Another important direction of our work is to control the outputs of deinterlacers. Sometimes, it can convert colorspace, work in BFF mode instead of TFF, or maybe have bugs.
Click on this text to read how exactly we validate deinterlacers' outputs.
The main criteria is that the PSNR between GT fields and the same fields in deinterlaced video must be equal to infinity. To control this, we make Top-Fields and Bottom-Fields plot on every second frame of GT video and deinterlaced video.
Here is the sample plot for Bob-Weave Deinterlacer. This deinterlacer passed the validation.
PSNR between Bottom-Fields of every deinterlaced frame and GT frame are equal to infinity, so we substitute infinity by zero. This means, that bottom field exactly matches the corresponding field in GT sequence.
Another sample plot, for MSU Deinterlacer, which had problems with colorspace.
As we can see here, Bottom-Field PSNR is about 100, but not infinity. In such cases, we make the following steps:
- Convert GT-data to the colorspace of the deinterlacer.
- If we don’t know (can’t guess) which colorspace the deinterlacer uses, we form the LookUp-Table from the fields of the deinterlacer output and GT data, that must be equal to each. Then, we use this LookUp-Table to map GT to the colorspace of the deinterlacer.
- In some cases, it is impossible to precisely determine LookUp-Table because the mapping is neither injective, nor surjective. In such hard cases, we just choose a GT sample with the highest PSNR from the previous 2 steps.
Also, we provide MSU Video Quality Measurement Tool (VQMT) PSNR visualization of the deinterlacer output. As the last step of the validation process, we check, that the VQMT PSNR visualization is striped. It should be so, because even/odds rows must exactly match the corresponding GT rows.
Here in the example of correct VQMT PSNR visualization output:
And, finally, let us make a closer look at it:
As we can see, the output is striped, so this means that the deinterlacer passed the last validation step.
There are 3 easy steps to submit:
- Download the interlaced video here.
We have more available formats, if YUV is not suitable. Just click on this text
There are 5 available options:
- a. Download frames of the video sequence in .png format here
- b. Download .yuv video file generated from frames via
ffmpeg -i %04d.png -c:v rawvideo -pix_fmt yuv444p sequences.yuvhere
- c. Download lossless encoded .mkv video generated from frames via
ffmpeg -i %04d.png -c:v libx264 -preset ultrafast -crf 0 -pix_fmt yuv444p lossless.mkvhere
- d. Download .rgb video file generated from frames via
ffmpeg -i %04d.png -c:v rawvideo -pix_fmt rgb24 sequences.rgbhere
- e. Download lossless encoded .avi video generated from frames via
ffmpeg -i %04d.png -c:v libx264rgb -preset ultrafast -crf 0 lossless.avihere
- Deinterlace downloaded video
The details, which may help you
TFF interlacing was used to get interlaced sequence from GT
The video consists of 40 videos, each separated by 5 black frames. Black frames are ignored
You can also send us any executable file or code of your deinterlacer, in order not to deinterlace video by yourself
- Send us an email to email@example.com
with the following information:
- A. Name of the deinterlacing method that will be specified in our benchmark
- B. Link to the cloud drive (Google Drive, OneDrive, Dropbox, etc.), containing deinterlaced
- C. (Optional) Any additional information about the method
Click here to see what may be included in additional information
- Technical information about deinterlaced video (e.g. colorspace, file-type, codec)
- The name of the theoretical method used
- Full name of the deinterlacing method or product
- The version that was used
- The parameter set that was used
- Any other additional information
- A link to the code of your deinterlacing method, if it is open-source
- A link to the paper about your deinterlacing method
- A link to the documentation of your deinterlacing method. For example, this is suitable for
deinterlacing methods that are implemented as a part of a video processing framework.
- A link to the page, where users can purchase or download your product (for example,
- D. (Optional) If you would like us to tune the parameters of your deinterlacing method, you
give us an ability to launch it. You can do it by sending us a code or an
executable file, providing us a free test version of your product, or any other possible way, that
is convenient for you
For questions and propositions, please contact us: firstname.lastname@example.org
- Codecs Comparison & Optimization
- Video Filters
- Video Quality Measurement Tool 3D
Semiautomatic Visual-Attention Modeling
MSU Benchmark Collection