This is part of MSU Video Quality Measurement tool Online Help for MSU VQMT 14.1
MSU Video Quality Measurement Tool (MSU VQMT) project:
- Main MSU VQMT page on compression.ru
- Download & Purchase on compression.ru
- Basic information
- Online help
- Metrics reference
- Version history & Changelog
Online help for MSU VQMT 14.1 contents:
- Command-line help
- VQMT various lists and tables
- Usage of VQMT metrics in CLI
- Picture types
- Configuration description
- PDF Documenation
- List of documentation pages for all versions
Content
List of available metrics
psnr
PSNR
https://videoprocessing.ai/vqmt/metrics/#psnr
- Color components: Y, U, V, R, G, B, LUV-L, RGB, YUV
- Type: reference metric
- Usage:
-metr psnr
[over <color component>]
identity
Identity
https://videoprocessing.ai/vqmt/metrics/#identity
- Color components: Y, U, V, R, G, B, LUV-L, RGB, YUV
- Type: reference metric
- Usage:
-metr identity
[over <color component>]
This metric can be configured using next parameter(s):
- Mode
Possible types:- binary - 1 images are similar, 0 in other case
- pixels - proportion of similar pixels, 1 - all pixels are same (0..1)
Default value:binary
Usage:-set "mode=
<value>"
, where <value> can be: binary
pixels
ssim
SSIM
https://videoprocessing.ai/vqmt/metrics/#ssim
- Color components: Y, U, V, R, G, B, LUV-L, RGB, YUV
- Type: reference metric
- Usage:
-metr ssim
[over <color component>]
This metric can be configured using next parameter(s):
- Combining mode
The mode of combining values for components of image:- default - in case of YUV image custom weight for Y component will be used, equal weights for U, V components will be used. In case if other color models equvivalent weights will be used
- ffmpeg - use area of component. Weights will depend on subsampling mode of image
Default value:default
Usage:-set "combining_mode=
<value>"
, where <value> can be: default
ffmpeg
- Y weight
If combining mode is default this is weight of value for component Y of YUV. Weights of U, V and other components is supposes to be 1.
Default value:4.0000000000000000
Usage:-set "y_weight=
<value>"
, where <value> can be:- 0.00000000000000000
- 999999.00000000000
- Usage:
-metr ssim_fast
[over <color component>] - Usage:
-metr ssim_precise
[over <color component>] - Usage:
-metr ssim_gpu_id
[over <color component>] - Color components: Y, U, V
- Usage:
-metr ssim_cuda
[over <color component>] - Color components: Y, U, V, R, G, B, LUV-L, RGB, YUV
- Usage:
-metr ssim
[over <color component>]-dev OpenCL0
This metric can be configured using next parameter(s):
- Combining mode
The mode of combining values for components of image:- default - in case of YUV image custom weight for Y component will be used, equal weights for U, V components will be used. In case if other color models equvivalent weights will be used
- ffmpeg - use area of component. Weights will depend on subsampling mode of image
Default value:default
Usage:-set "combining_mode=
<value>"
, where <value> can be: default
ffmpeg
- Y weight
If combining mode is default this is weight of value for component Y of YUV. Weights of U, V and other components is supposes to be 1.
Default value:4.0000000000000000
Usage:-set "y_weight=
<value>"
, where <value> can be:- 0.00000000000000000
- 999999.00000000000
msssim
MS-SSIM
https://videoprocessing.ai/vqmt/metrics/#msssim
- Color components: Y, U, V, R, G, B, LUV-L, RGB, YUV
- Type: reference metric
- Usage:
-metr msssim
[over <color component>]
This metric can be configured using next parameter(s):
- Combining mode
The mode of combining values for components of image:- default - in case of YUV image custom weight for Y component will be used, equal weights for U, V components will be used. In case if other color models equvivalent weights will be used
- ffmpeg - use area of component. Weights will depend on subsampling mode of image
Default value:default
Usage:-set "combining_mode=
<value>"
, where <value> can be: default
ffmpeg
- Y weight
If combining mode is default this is weight of value for component Y of YUV. Weights of U, V and other components is supposes to be 1.
Default value:4.0000000000000000
Usage:-set "y_weight=
<value>"
, where <value> can be:- 0.00000000000000000
- 999999.00000000000
https://videoprocessing.ai/vqmt/metrics/#ssim
- Usage:
-metr msssim_fast
[over <color component>]
https://videoprocessing.ai/vqmt/metrics/#msssim - Usage:
-metr msssim_precise
[over <color component>] - Color components: Y, U, V
- Usage:
-metr msssim_cuda
[over <color component>] - Color components: Y, U, V, R, G, B, LUV-L, RGB, YUV
- Usage:
-metr msssim
[over <color component>]-dev OpenCL0
This metric can be configured using next parameter(s):
- Combining mode
The mode of combining values for components of image:- default - in case of YUV image custom weight for Y component will be used, equal weights for U, V components will be used. In case if other color models equvivalent weights will be used
- ffmpeg - use area of component. Weights will depend on subsampling mode of image
Default value:default
Usage:-set "combining_mode=
<value>"
, where <value> can be: default
ffmpeg
- Y weight
If combining mode is default this is weight of value for component Y of YUV. Weights of U, V and other components is supposes to be 1.
Default value:4.0000000000000000
Usage:-set "y_weight=
<value>"
, where <value> can be:- 0.00000000000000000
- 999999.00000000000
3ssim
3SSIM
https://videoprocessing.ai/vqmt/metrics/#3ssim
- Color components: Y, U, V
- Type: reference metric
- Usage:
-metr 3ssim
[over <color component>] - Usage:
-metr 3ssim_cuda
[over <color component>] - Usage:
-metr 3ssim
[over <color component>]-dev OpenCL0
vqm
VQM
https://videoprocessing.ai/vqmt/metrics/#vqm
- Color components: Y
- Type: reference metric
- Usage:
-metr vqm
[over <color component>]
blocking
Blocking
https://videoprocessing.ai/vqmt/metrics/#blockingmeasure
- Color components: Y
- Type: no-reference metric
- Usage:
-metr blocking
[over <color component>]
blurring
Blurring
https://videoprocessing.ai/vqmt/metrics/#ybluringmeasure
- Color components: Y, R, G, B
- Type: no-reference metric
- Usage:
-metr blurring
[over <color component>] - Color components: Y, U, V, R, G, B
- Usage:
-metr blurring_delta
[over <color component>]
delta
Delta
https://videoprocessing.ai/vqmt/metrics/#delta
- Color components: Y, U, V, R, G, B, LUV-L
- Type: reference metric
- Usage:
-metr delta
[over <color component>]
msad
MSAD
https://videoprocessing.ai/vqmt/metrics/#msad
- Color components: Y, U, V, R, G, B, LUV-L
- Type: reference metric
- Usage:
-metr msad
[over <color component>]
mse
MSE
https://videoprocessing.ai/vqmt/metrics/#mse
- Color components: Y, U, V, R, G, B, LUV-L
- Type: reference metric
- Usage:
-metr mse
[over <color component>]
time-shift
Time shift
https://videoprocessing.ai/vqmt/metrics/#shift
- Color components: Y
- Type: reference metric
- Usage:
-metr time-shift
[over <color component>]
This metric can be configured using next parameter(s):
- Max. shift
Maximum shift, that can be detected. Note: large values leads big memory consumption
Default value:5
Usage:-set "max-shift=
<value>"
, where <value> can be:- value in range 0..25
- Direction
Detect only positive shifts (frame dups), negatives (frame drops) or both
Default value:both
Usage:-set "direction=
<value>"
, where <value> can be:positive
negative
both
- Destination metric
This metric will be used to measure similarity between frames
Default value:psnr
Usage:-set "metric=
<value>"
, where <value> can be:psnr
ssim
- Show metric values
Metric will output now only shift, but destination metric values
Default value:false
Usage:-set "show-metric=
<value>"
, where <value> can be:true
false
- Threshold
We will consider shift only if metric for neighbour frame better than this thresold multiplied to metric for similar frame
Default value:0.99500000476837158
Usage:-set "threshold=
<value>"
, where <value> can be:- any floating point number
- Smoothing
Will smooth metric values over time. If equal n, than smoothing will be in the interval frame-n..frame+n
Default value:1
Usage:-set "smoothing=
<value>"
, where <value> can be:- value in range 0..25
si
SI / TI
https://videoprocessing.ai/vqmt/metrics/#si
- Color components: Y
- Type: no-reference metric
- Usage:
-metr si
[over <color component>]-dev CPU
ti
SI / TI
https://videoprocessing.ai/vqmt/metrics/#ti
- Color components: Y
- Type: no-reference metric
- Usage:
-metr ti
[over <color component>]-dev CPU
niqe
NIQE
https://videoprocessing.ai/vqmt/metrics/#niqe
- Color components: Y
- Type: no-reference metric
- Usage:
-metr niqe
[over <color component>]
This metric can be configured using next parameter(s):
- Mean threshold
Values of metric greater than this value will be skipped during mean calculation. 0 for disable skipping
Default value:15.000000000000000
Usage:-set "mean_thresh=
<value>"
, where <value> can be:- any floating point number
- Threshold smoothing
Values of metric greater than ‘Mean threshold’ + ‘Threshold smoothing’ will be skipped, values less than ‘Mean threshold’ - ‘Threshold smoothing’ will be assumed with weight 1. Intermediate values will be taken with intermediate weight
Default value:5.0000000000000000
Usage:-set "mean_thresh_smoothing=
<value>"
, where <value> can be:- any floating point number
- Type of normalization
- fast - the fastest algorithm, low precision
- native - like in native NIQE implementation. Slowest one
- precise - the most precise algorithm
Default value:native
Usage:-set "norm_alg=
<value>"
, where <value> can be:
fast
native
precise
- fast - the fastest algorithm, low precision
- Usage:
-metr niqe
[over <color component>]-dev OpenCL0
This metric can be configured using next parameter(s):
- Mean threshold
Values of metric greater than this value will be skipped during mean calculation. 0 for disable skipping
Default value:15.000000000000000
Usage:-set "mean_thresh=
<value>"
, where <value> can be:- any floating point number
- Threshold smoothing
Values of metric greater than ‘Mean threshold’ + ‘Threshold smoothing’ will be skipped, values less than ‘Mean threshold’ - ‘Threshold smoothing’ will be assumed with weight 1. Intermediate values will be taken with intermediate weight
Default value:5.0000000000000000
Usage:-set "mean_thresh_smoothing=
<value>"
, where <value> can be:- any floating point number
vmaf
Netflix VMAF
- Color components: Y
- Type: reference metric
- Usage:
-metr vmaf
[over <color component>]
This metric can be configured using next parameter(s):
- Model preset
Choose built-in model or ‘custom’ for loading model from file. Built-in models:- default - VMAF default behaviour:
- VMAF v0.6.1 for running without confidence interval and per-model values
- VMAF v0.6.1 4k for previous case if applying 4k model
- VMAF v0.6.3 for running with confidence interval or per-model values
- VMAF v0.6.2 4k for previous case if applying 4k model (NOTE: no v0.6.3 for 4k)
- vmaf_v061 - Netflix model VMAF v0.6.1 (2k or 4k)
- vmaf_v061_neg - Netflix model VMAF v0.6.1 (only 2k), no enhancement gain
- vmaf_v062 - Netflix model VMAF v0.6.2 (2k or 4k), supports confidence interval
- vmaf_v063 - Netflix model VMAF v0.6.3 (only 2k), supports confidence interval
- all_models - vmaf_v061..vmaf_v063, vmaf_v061_neg computed sumultaneously
- basic_features - view only basic features from VMAF. Model will not be applied
- standard_features - features that is used in VMAF v0.6.1 and VMAF score (2k or 4k)
- all_features - view all features from VMAF. Model will not be applied
- all_features_with_neg - view all features from VMAF and no enhancement gain features. Model will not be applied
- all - all feature and next models:
- VMAF v0.6.1 (2k or 4k)
- VMAF v0.6.1 no enhancement gain (neg, 2k only)
- VMAF v0.6.2 (2k or 4k)
- VMAF v0.6.3
Default value:default
Usage:-set "model_preset=
<value>"
, where <value> can be:
default
vmaf_v061
vmaf_v061_neg
vmaf_v062
vmaf_v063
vmaf_v060
all_models
basic_features
standard_features
standard_features_neg
all_features
all_features_with_neg
all
custom
- default - VMAF default behaviour:
- Custom model (*.pkl or JSON file)
you can specify path to *.pkl or JSON file here (or multiple ;-separated *.pkl or JSON files). Model file should be placed near PKL.
NOTE: this only means if preset is set to ‘custom’
Default value: ``
Usage:-set "custom_model_files=
<value>"
, where <value> can be:- any string
- 4k
selection 4k model policy:- auto - select 4k if exists suitable model and input video is 4k
- forced_2k - always 2k model
- forced_4k - 4k if exsists: VMAF v0.6.1-2
NOTE: this param does not affects custom model
Default value:auto
Usage:-set "4k=
<value>"
, where <value> can be: auto
forced_2k
forced_4k
- Confidence interval
turn on additional VMAF features: 95%-confidence interval output and other statistical information
Default value:false
Usage:-set "confidence_interval=
<value>"
, where <value> can be:true
false
- Confidence interval size
the length of confidence interval if turned on, percent
Default value:95.000000000000000
Usage:-set "ci_size=
<value>"
, where <value> can be:- any floating point number
- Per-model values
output values for all bootstrap models if confidence interval is on
Default value:false
Usage:-set "permodel_values=
<value>"
, where <value> can be:true
false
- Bootstrap type
output values for all bootstrap models if confidence interval is on
Default value:common
Usage:-set "bootstrap_type=
<value>"
, where <value> can be:common
residue
- Visualize algorithm (if on)
if visualization turned on you can select feature to visualize. It’s impossible to calculate distribution of real VMAF value, so you can only visualize one of supposed features
Default value:default
Usage:-set "visualize_alg=
<value>"
, where <value> can be:default
adm
ansnr
motion
vif
adm
motion
vif
- Use phone model
turn on postprocessing of metric value that produces more precise results for handheld devices. Select ‘both’ to see both results with and without postprocessing
Default value:no
Usage:-set "phone_model=
<value>"
, where <value> can be:no
yes
both
- Disable clipping values
turn off clipping value to range set by model (0..100 for example)
Default value:false
Usage:-set "disable_clip=
<value>"
, where <value> can be:true
false
- Model internal datatype (integer or float)
turn off clipping value to range set by model (0..100 for example)
Default value:float
Usage:-set "datatype=
<value>"
, where <value> can be:float
integer
- Usage:
-metr vmaf
[over <color component>]-dev OpenCL0
vmaf_legacy
Netflix VMAF legacy
- Color components: Y
- Type: reference metric
- Usage:
-metr vmaf_legacy
[over <color component>]
This metric can be configured using next parameter(s):
- Model preset
Choose built-in model or ‘custom’ for loading model from file. Built-in models:- default - VMAF default behaviour:
- VMAF v0.6.1 for running without confidence interval and per-model values
- VMAF v0.6.1 4k for previous case if applying 4k model
- VMAF v0.6.3 for running with confidence interval or per-model values
- VMAF v0.6.2 4k for previous case if applying 4k model (NOTE: no v0.6.3 for 4k)
- vmaf_v061 - Netflix model VMAF v0.6.1 (2k or 4k)
- vmaf_v061_neg - Netflix model VMAF v0.6.1 (only 2k), no enhancement gain
- vmaf_v062 - Netflix model VMAF v0.6.2 (2k or 4k), supports confidence interval
- vmaf_v063 - Netflix model VMAF v0.6.3 (only 2k), supports confidence interval
- all_models - vmaf_v061..vmaf_v063, vmaf_v061_neg computed sumultaneously
- basic_features - view only basic features from VMAF. Model will not be applied
- standard_features - features that is used in VMAF v0.6.1 and VMAF score (2k or 4k)
- all_features - view all features from VMAF. Model will not be applied
- all_features_with_neg - view all features from VMAF and no enhancement gain features. Model will not be applied
- all - all feature and next models:
- VMAF v0.6.1 (2k or 4k)
- VMAF v0.6.1 no enhancement gain (neg, 2k only)
- VMAF v0.6.2 (2k or 4k)
- VMAF v0.6.3
Default value:default
Usage:-set "model_preset=
<value>"
, where <value> can be:
default
vmaf_v061
vmaf_v062
vmaf_v063
vmaf_v060
all_models
basic_features
standard_features
all_features
all
custom
- default - VMAF default behaviour:
- Custom model (*.pkl or JSON file)
you can specify path to *.pkl or JSON file here (or multiple ;-separated *.pkl or JSON files). Model file should be placed near PKL.
NOTE: this only means if preset is set to ‘custom’
Default value: ``
Usage:-set "custom_model_files=
<value>"
, where <value> can be:- any string
- 4k
selection 4k model policy:- auto - select 4k if exists suitable model and input video is 4k
- forced_2k - always 2k model
- forced_4k - 4k if exsists: VMAF v0.6.1-2
NOTE: this param does not affects custom model
Default value:auto
Usage:-set "4k=
<value>"
, where <value> can be: auto
forced_2k
forced_4k
- Confidence interval
turn on additional VMAF features: 95%-confidence interval output and other statistical information
Default value:false
Usage:-set "confidence_interval=
<value>"
, where <value> can be:true
false
- Confidence interval size
the length of confidence interval if turned on, percent
Default value:95.000000000000000
Usage:-set "ci_size=
<value>"
, where <value> can be:- any floating point number
- Per-model values
output values for all bootstrap models if confidence interval is on
Default value:false
Usage:-set "permodel_values=
<value>"
, where <value> can be:true
false
- Bootstrap type
output values for all bootstrap models if confidence interval is on
Default value:common
Usage:-set "bootstrap_type=
<value>"
, where <value> can be:common
residue
- Visualize algorithm (if on)
if visualization turned on you can select feature to visualize. It’s impossible to calculate distribution of real VMAF value, so you can only visualize one of supposed features
Default value:default
Usage:-set "visualize_alg=
<value>"
, where <value> can be:default
adm
ansnr
motion
vif
adm
motion
vif
- Use phone model
turn on postprocessing of metric value that produces more precise results for handheld devices. Select ‘both’ to see both results with and without postprocessing
Default value:no
Usage:-set "phone_model=
<value>"
, where <value> can be:no
yes
both
- Disable clipping values
turn off clipping value to range set by model (0..100 for example)
Default value:false
Usage:-set "disable_clip=
<value>"
, where <value> can be:true
false
- Model internal datatype (integer or float)
turn off clipping value to range set by model (0..100 for example)
Default value:float
Usage:-set "datatype=
<value>"
, where <value> can be:float
integer
- Usage:
-metr vmaf_legacy
[over <color component>]-dev OpenCL0
hdr-psnr
PSNR
https://videoprocessing.ai/vqmt/metrics/#psnr
- Color components: PU-L, PU encoding BT.709 L, PU encoding BT.2020 L, ICtCp BT.2100 PQ ICtCp
- Type: reference metric
- Usage:
-metr hdr-psnr
[over <color component>]
This metric can be configured using next parameter(s):
- Peak luminance (nits)
PSNR peak luminance value. 0 means display luminance
Default value:0.00000000000000000
Usage:-set "peak_lum=
<value>"
, where <value> can be:- value in range 0..10000
hdr-ssim
HDR SSIM
https://videoprocessing.ai/vqmt/metrics/#ssim
- Color components: PU-L, PU encoding BT.709 L, PU encoding BT.2020 L
- Type: reference metric
- Usage:
-metr hdr-ssim
[over <color component>]-dev CPU
This metric can be configured using next parameter(s):
- Combining mode
The mode of combining values for components of image:- default - in case of YUV image custom weight for Y component will be used, equal weights for U, V components will be used. In case if other color models equvivalent weights will be used
- ffmpeg - use area of component. Weights will depend on subsampling mode of image
Default value:default
Usage:-set "combining_mode=
<value>"
, where <value> can be: default
ffmpeg
- Y weight
If combining mode is default this is weight of value for component Y of YUV. Weights of U, V and other components is supposes to be 1.
Default value:4.0000000000000000
Usage:-set "y_weight=
<value>"
, where <value> can be:- 0.00000000000000000
- 999999.00000000000
- Usage:
-metr hdr-ssim
[over <color component>]-dev CPU
- Usage:
-metr hdr-ssim
[over <color component>]-dev CPU
- Usage:
-metr hdr-ssim
[over <color component>]-dev OpenCL0
hdr-msssim
HDR MS-SSIM
https://videoprocessing.ai/vqmt/metrics/#msssim
- Color components: PU-L, PU encoding BT.709 L, PU encoding BT.2020 L
- Type: reference metric
- Usage:
-metr hdr-msssim
[over <color component>]-dev CPU
This metric can be configured using next parameter(s):
- Combining mode
The mode of combining values for components of image:- default - in case of YUV image custom weight for Y component will be used, equal weights for U, V components will be used. In case if other color models equvivalent weights will be used
- ffmpeg - use area of component. Weights will depend on subsampling mode of image
Default value:default
Usage:-set "combining_mode=
<value>"
, where <value> can be: default
ffmpeg
- Y weight
If combining mode is default this is weight of value for component Y of YUV. Weights of U, V and other components is supposes to be 1.
Default value:4.0000000000000000
Usage:-set "y_weight=
<value>"
, where <value> can be:- 0.00000000000000000
- 999999.00000000000
https://videoprocessing.ai/vqmt/metrics/#ssim
- Usage:
-metr hdr-msssim
[over <color component>]-dev CPU
https://videoprocessing.ai/vqmt/metrics/#msssim - Usage:
-metr hdr-msssim
[over <color component>]-dev CPU
- Usage:
-metr hdr-msssim
[over <color component>]-dev OpenCL0
This metric can be configured using next parameter(s):
- Combining mode
The mode of combining values for components of image:- default - in case of YUV image custom weight for Y component will be used, equal weights for U, V components will be used. In case if other color models equvivalent weights will be used
- ffmpeg - use area of component. Weights will depend on subsampling mode of image
Default value:default
Usage:-set "combining_mode=
<value>"
, where <value> can be: default
ffmpeg
- Y weight
If combining mode is default this is weight of value for component Y of YUV. Weights of U, V and other components is supposes to be 1.
Default value:4.0000000000000000
Usage:-set "y_weight=
<value>"
, where <value> can be:- 0.00000000000000000
- 999999.00000000000
hdr-vqm
HDRVQM
https://sites.google.com/site/narwariam/home/research/hdr-vqm
- Color components: PU-L, PU encoding BT.709 L, PU encoding BT.2020 L
- Type: reference metric
- Usage:
-metr hdr-vqm
[over <color component>]
This metric can be configured using next parameter(s):
- Fixation frames
Number of frame for calculation of spatio-temporal tubes. 0 - auto (0.6 seconds, based on fps of first video)
Default value:0
Usage:-set "fixation_frames=
<value>"
, where <value> can be:- value in range 0..50
- FFT cols
Video will be scaled to this value (bicubic). 0 - auto (clothest to video size)
Default value:1024
Usage:-set "fft_cols=
<value>"
, where <value> can be:- value in range 64..32768
- FFT rows
Video will be scaled to this value (bicubic). 0 - auto (clothest to video size)
Default value:512
Usage:-set "fft_rows=
<value>"
, where <value> can be:- value in range 64..32768
- Pooling Percentage
Pooling Percentage for long-term pooling
Default value:0.30000001192092896
Usage:-set "pooling_precent=
<value>"
, where <value> can be:- value in range 0.10000000000000001..1.0000000000000000
- Dispaly: cols
Columns of the target display
Default value:1920.0000000000000
Usage:-set "display_cols=
<value>"
, where <value> can be:- value in range 500..10000
- Dispaly: rows
Rows of the target display
Default value:1080.0000000000000
Usage:-set "display_rows=
<value>"
, where <value> can be:- value in range 500..10000
- Dispaly: area
Area of the target display
Default value:6100.0000000000000
Usage:-set "display_area=
<value>"
, where <value> can be:- value in range 500.00000000000000..100000.00000000000
- Dispaly: distance
Viewer’s dispaly distance, cm
Default value:178.00000000000000
Usage:-set "display_distance=
<value>"
, where <value> can be:- value in range 10.000000000000000..100.00000000000000
delta-ictcp
Delta
https://videoprocessing.ai/vqmt/metrics/#delta
- Color components: ICtCp BT.2100 PQ ICtCp
- Type: reference metric
- Usage:
-metr delta-ictcp
[over <color component>]
This page is automatically generated by 14.1 r12839 on 2022-06-24. In case of any question or suggestion, please mail us: video-measure@compression.ru
-
MSU Benchmark Collection
- Video Colorization Benchmark
- Super-Resolution for Video Compression Benchmark
- Defenses for Image Quality Metrics Benchmark
- Learning-Based Image Compression Benchmark
- Super-Resolution Quality Metrics Benchmark
- Video Saliency Prediction Benchmark
- Metrics Robustness Benchmark
- Video Upscalers Benchmark
- Video Deblurring Benchmark
- Video Frame Interpolation Benchmark
- HDR Video Reconstruction Benchmark
- No-Reference Video Quality Metrics Benchmark
- Full-Reference Video Quality Metrics Benchmark
- Video Alignment and Retrieval Benchmark
- Mobile Video Codecs Benchmark
- Video Super-Resolution Benchmark
- Shot Boundary Detection Benchmark
- The VideoMatting Project
- Video Completion
- Codecs Comparisons & Optimization
- VQMT
- MSU Datasets Collection
- Metrics Research
- Video Quality Measurement Tool 3D
- Video Filters
- Other Projects