This is a legacy SmoothVideo Project (SVP) ver.3 user manual.
Active FRC profile settings window opens by double click on the SVP icon in the notification area or by selecting one of the profiles in submenu of Video profiles.
Profiles describe the set of parameters for the FRC algorithm and are selected automatically depending on the video played, refer to Managing profiles for details.
The set of parameters in the configuration window is defined by selected interface type (basic, advanced или expert).
Attention: editing profiles, especially using interface type higher than "basic", requires special knowledge. For instance setting "blindly" all setting to the "maximum" will not achieve a "maximal" result.
SVP already includes a large number of pre-built profiles covering the majority of video formats. Start editing a profile manually only if the default profiles for some reason do not fit your needs.
The optimal parameter settings (from SVP Team point of view) independent from the frame size or selected frame rate are marked with "default".
It is important to understand that for any parameters the balance between video smoothness and quality / noticeable artifacts FRC is kept. So high smoothness always leads to stronger artifacts and artifacts reduction always reduces smoothness.
There is no perfect set of parameters which gives the maximum smoothness without artifacts.
Almost all FRC profile parameters are directly passed to SVPflow plugin functions to perform the basic calculations, hence experts can immediately jump to SVPflow parameters description.
- 1 Basic options
- 2 Advanced options
- 3 Expert options
The "basic" interface contains only the necessary parameters, which can not significantly damage the result.
Frames interpolation mode
- Off - disables the FRC for current profile. Can be useful for creating special profiles for video format UHD 4K, or hardware backlight SVPlight
- Uniform - the most "honest" mode in which the maximum smoothness for the specified target frame rate is achieved. The number of frames of the original video, in the output is minimal that gives the most visibility of artifacts.
The following options retain all the original frames, and differ in the amount and uniformity of interpolated frames:
- 2m (every original frame twice) - the largest number of repeats the original frame, minimal smoothness
- 1m (every original frame once) - each original frame is repeated once
- 1.5m (like 1m but with different intermediate frames times)
More details about "-m" modes can be read in Technical details section.
- Adaptive - automatically switches modes on the basis of the analysis of the resulting quality of FRC for a given scene
Specifies an algorithm for constructing an intermediate frame of a pair of source and found motion vectors. It does not make sense to describe the algorithms in the manual, but interested individuals can refer to SVPflow plugins.
- 1. Fastest - the fastest :) and sharpest (does not mix objects)
- 2. Sharp - does not mix objects, less artifacts than with 1st shader, recommended for hand-drawn animation.
- 11. Simple Lite - most simple mixing of compensated frame in motion
- 11b. By blocks - same as 11, but in block mode, much faster with disabled GPU-acceleration, but gives noticeable square artifacts
- 13. Standard - variation of 11th shader, giving more visible artifacts but more "soaping" effect
- 21. Simple - 11th shader, supplemented with masking
- 23. Complicated - 21st shader with more complex masking
Target frame rate
Selection of target frame rate.
Important: For smooth video playback, without jerks and fading, video frame rate should be close to the vertical frequency of the monitor.
The basic parameters offer two options:
- Double source frame rate (2х) - simple doubling of the frame rate of the original video gives the fewest FRC artifacts, but there may be jerks because of the video frame rate not fitting with a vertical refresh rate of the monitor.
- To screen refresh rate - smoothing up to the vertical refresh rate of the monitor detected by the Manager, or set manually by menu item Screen selection -> Force....
Decrease frame size
- To screen size
The speed of the FRC algorithm depends on the size of a video frame.
Therefore, when a video is playing more than the size of the monitor one from the two options:
- Video -> FRC -> decrease the screen
- Video -> reduction of the screen -> FRC -> output point-to-point
the second should be chosen.
If there is insufficient capacity of the system (especially CPU) it is possible as last resort to reduce the whoe video before FRC, reducing the resolution and video definition:
- To 50% of original size
- To 75% of original size
Important: using this parameter changes the frame size that can be handled incorrectly by some video players.
Target frame rate
In addition to the basic parameters there are the following values with fixed frame rate increasing factors:
- 25->60 (2.4x)
- 24->60 (2.5x)
- Source multiplied by 3
Attention: the output frame rate higher then the vertical frame rate of the monitor will lead to jerks and fading of the video
We recommend to us always the To screen refresh rate mode (from the basic parameters).
Motion vectors grid
Selected FRC algorithm searches for motion in the frame on small image blocks. The smaller block size is, the:
- More accurate is the determination of local movements
- Significantly higher are the CPU requirements
- More noticeable wavy artifacts
Blocks can overlap each other by a quarter or a half unit. Overlapping increases requirements for CPU, but always increases quality.
All together, this gives nine versions of grid stepping vectors (three sizes for three overlap variants):
- Large 1/2/3
- Average 1/2/3
- Small 1/2/3
Important: small block size is not always the best in terms of the final result, even if the capacity of the system allows you to set a block size. We recommend to set Average block size with average overlapping.
More to the topic of the block size can be found in MVTools description.
Motion vectors precision
The higher the accuracy, the higher the smoothness of slow movements but CPU and RAM requirements are also higher, especially without GPU-acceleration.
- Two pixels
- One pixel
- Half pixel
It is important to understand that any smoothness has a limit, and if for DVD the difference between FRC modes may be noticeable immediately, for FullHD the difference is clearly visible only in rare cases which are especially selected, then for video format 4K may not be any difference at all.
Masking "bad" areas of the frame (i.e. areas in which the motion is determined with large errors, and where will be more visible artifacts) parts of the original frame.
Weak masking usually gives good results, but it can lead to new types of artifacts, such as tripling of thin objects. Strong masking further significantly reduces the overall visual smoothness of the video.
- Weakest ... Average ... Strongest
Processing of scene changes
Behavior on "change of scene", when one frame shows one scene in a movie, and the following frame switches to another.
A good example - an ordinary dialogue, the camera alternates between the two people. At scene change it is senseless to search movement since there is none. Moreover, an attempt to find such a motion, and build intermediate frames causes artifacts as large as the size of the entire frame. Therefore SVP specifically looks for moments of change of scene and processes them in a special way.
- Blend adjacent frames - smooth transition from one scene to another by mixing (reconstitution) frames between each other
- Repeat frame - repeat the original frame, the scene changes instantly
Decrease frame size
Additional to basiс parameters it is possible to decrease the frame size to [specified] width/height.
The influence of these parameters on the results has not been understood by ourselves :)
Secret knowledge seekers can start by exploring help of MVTools, continue reading on Doom9 forums, and also take part in our forum.
Target frame rate
In addition to the advanced settings the following values are available:
- To screen refresh rate /3
- To screen refresh rate /2
Decrease grid step
- To small step 6-8 px.
- To smallest step 3-4 px.
- By two with local refinement
- By two with global refinement
When reducing the grid, the FRC algorithm tries to clarify the motion vector with smallest block size. Unfortunately, there are no people in this world that know for 100% how the verification procedure will affect a particular video, hence we drop the details in this Article.
Limits the maximum length of motion vectors. Large radius is not always better than small, because the found "far" motion vector may be incorrect, resulting in more artifacts. The larger the search radius the higher is the demand on CPU.
Also, this setting specifies the method for comparing image blocks together, all options except Small and fast are using the SADT function instead of SAD.
- Small and fast
- Small / Average / Large
What to do if a good enough motion vector couldn't be found at a given search radius.
- Average / Strong / Strongest