TFM

Abstract

• YV12 & YUY2 Colorspace

license: GPL

Table of Contents

• A.) General Info
• B.) Basic Parameters
• C.) Advanced Parameters
• D.) File Input/Output Parameters
• E.) Debug/Display Parameters
• F.) Overrides Info
• G.) Changelist

A.) GENERAL INFO:

TFM is a field matching filter that recovers the original progressive frames from a telecined stream. It does not decimate the resulting duplicate frames though, so to achieve an ivtc you must follow TFM with a decimation filter such as TDecimate(), which is also included in the tivtc.dll.

syntax

TFM (clip, int "order", int "field", int "mode", int "PP". string "ovr", string "input", string "output", string "outputC", bool "debug", bool "display", bool "slow", bool "mChroma", int "cNum", int "cthresh", int "MI", bool "chroma", int "blockx", int "blocky", int "y0", int "y1", int "mthresh", PClip "clip2", string "d2v", int "ovrDefault")

While TFM does have quite a few parameters, I have tried to categorize the settings so they are easier to manage. Really there are only 3 parameters most users should be concerned about knowing or ever setting. They are "order", "mode", and "PP" and are described under the "BASIC PARAMETERS" sections. The other settings are described under the other various sections which should be pretty easy to figure out from the names.

B.) BASIC PARAMETERS:

These are the only three parameters most users should be concerned worth or worry about setting. They control the basic operation of the filter.

order - 

Sets the field order of the clip. The field order must be correctly set for accurate field matching to occur. Possible values:

-1 = auto (use avisynth's internal parity value)
0 = bottom field first (bff)
1 = top field first (tff)

Default: -1 (int)

mode - 

Sets the matching mode or strategy to use. Plain 2-way matching (option 0) is the safest of all the options in the sense that it wont risk creating jerkiness due to duplicate frames when possible, but if there are bad edits or blended fields it will end up outputting combed frames when a good match might actually exist. 3-way matching + trying the 4th/5th matches if all 3 of the original matches are detected as combed (option 5) is the most risky in terms of creating jerkiness, but will almost always find a good frame if there is one. The other settings (options 1, 2, 3, and 4) are all somewhere in between options 0 and 5 in terms of risking jerkiness and creating duplicate frames vs. finding good matches in sections with bad edits, orphaned fields, blended fields, etc... Possible settings are:

0 = 2-way match (p/c)
1 = 2-way match + 3rd match on combed (p/c + n)
2 = 2-way match + 3rd match (same order) on combed (p/c + u)
3 = 2-way match + 3rd match on combed + 4th/5th matches if still combed (p/c + n + u/b)
4 = 3-way match (p/c/n)
5 = 3-way match + 4th/5th matches on combed (p/c/n + u/b)

The () at the end indicate the matches that would be used for that mode assuming order = 1 and field = 1.

* In terms of speed option 0 is by far the fastest and 5 is the slowest.

Default: 1 (int)

PP -

Sets the post-processing mode. This controls how TFM should handle (or not handle) any combed frames that come out of the field matching process. Possible options are:

0 - nothing (don't even look for combed frames)
1 - find/hint combed frames but don't deinterlace
2 - dumb blend deinterlacing
3 - dumb cubic interpolation deinterlacing
4 - dumb modified-ela deinterlacing
5 - motion-adaptive blend deinterlacing
6 - motion-adaptive cubic interpolation deinterlacing
7 - motion-adaptive modified-ela deinterlacing

* You can also use the "clip2" parameter to specify an externally deinterlaced clip from which TFM will take frames instead of doing the deinterlacing itself. See the clip2 parameter description for more info and to see how the PP settings effect operation with clip2.

* You can manually control what frames should be marked as combed as well as the PP option through an overrides file (see the ovr parameter description)

Default: 6 (int)

C.) ADVANCED PARAMETERS:

These parameters give finer control over the matching operation, combed frame detection, and deinterlacing of combed frames. While correctly tweaking these values can increase
matching accuracy they can also make things worse if just changed randomly.

field - 

Sets the field to match from. It is recommended to set this to the same value as order unless you experience matching failures with that setting. In certain circumstances changing the field that is used to match from can have a large impact on matching performance. Possible values are:

-1 = auto (TFM will set field to the same value as the order parameter)
0 = bottom field
1 = top field

Default: -1 (int)

slow -

Sets whether or not the slower matching function or the faster matching function is used. The slower version is usually more accurate, but as the name says is slower. This is basically a quality vs. speed option.

true = slower version is used
false = faster version is used

Default: true (bool)

mChroma -

Sets whether or not chroma is included during the match comparisons. In most cases it is recommended to leave this enabled. Only if your clip has bad chroma problems such as heavy rainbowing or other artifacts should you set this to false. Setting this to false could also be used to speed things up at the cost of some accuracy.

true = chroma is included
false = chroma is not included

Default: true (bool)

y0/y1 -

These define an exclusion band which excludes the lines between y0 and y1 from being included in the field matching decision. An exclusion band can be used to ignore subtitles, a logo, or other things that may interfer with the matching. y0 sets the starting scan line and y1 sets the ending line, all lines in between y0 and y1 (including y0 and y1) will be ignored. Set y0 equal to y1 to disable.

Default: y0 - 0 (int)
y1 - 0 (int)

The rest of the "advanced parameters" control combed frame detection and deinterlacing, and are only used if mode is set to 1, 2, or 4 or PP (post-processing) is set to 1 or greater!

cthresh -

This is the area combing threshold used for combed frame detection. It is like dthresh or dthreshold in telecide() and fielddeinterlace(). This essentially controls how "strong" or "visible" combing must be to be detected. Larger values mean combing must be more visible and smaller values mean combing can be less visible or strong and still be detected. Valid settings are from -1 (every pixel will be detected as combed) to 255 (no pixel will be detected as combed). This is bascially a pixel difference value. A good range is between 8 to 12.

Default: 10 (int)

chroma -

Sets whether or not chroma is considered in the combed frame decision. Only disable this if your source has chroma problems (rainbowing, etc...) that are causing problems for the combed frame detection with chroma enabled.

true = chroma is included
false = chroma is not included

Default: false (bool)

blockx -

Sets the x-axis size of the window used during combed frame detection. This has to do with the size of the area in which MI number of pixels are required to be detected as combed for a frame to be declared combed. See the MI parameter description for more info. Possible values are any number that is a power of 2 starting at 4 and going to 2048 (i.e. 4, 8, 16, 32, ... 2048).

Default: 16 (int)

blocky -

Sets the y-axis size of the window used during combed frame detection. This has to do with the size of the area in which MI number of pixels are required to be detected as combed for a frame to be declared combed. See the MI parameter description for more info. Possible values are any number that is a power of 2 starting at 4 and going to 2048 (i.e. 4, 8, 16, 32, ... 2048).

Default: 16 (int)

MI -

The # of combed pixels inside any of the blocky by blockx size blocks on the frame for the frame to be detected as combed. While cthresh controls how "visible" the combing must be, this setting controls "how much" combing there must be in any localized area (a window defined by the blockx and blocky settings) on the frame. Min setting = 0, max setting = blocky x blockx (at which point no frames will ever be detected as combed).

Default: 85 (int)

mthresh - 

Sets the motion (pixel difference) threshold for deinterlacing when using motion adaptation (PP > 4). As said, it is simply a pixel difference threshold between frames... if under mthresh then a pixel is considered static, if over mthresh it is considered moving.

Default: 5 (int)

clip2 -

When using PP > 1 this parameter may be used to specify a clip to take deinterlaced frames from instead of using one of TFM's inbuilt deinterlacing methods. If PP < 5 then frames will be taken as is from clip2, if PP >= 5 (i.e. using motion adaptation) then TFM will build the mask as usual and only pixels in moving areas will be taken from the frames in clip2. Don't specify anything for clip2 to use TFM's internal deinterlacing. For an example of using clip2 see the clip2_sample_avs.avs file.

Default: NULL (PClip)

D.) FILE INPUT/OUTPUT PARAMETERS:

These settings deal with overrides file usage and output/input file options.

d2v - 

This option is intended to be used if you are using an mpeg2source() with a d2v file. It sets the name and path to a d2v file, which TFM will analyze to see if there are any illegal field order changes and optionally set the order parameter using the field order of the d2v file. If the d2v file is found to have illegal field order transitions, TFM will create a fixed d2v file with the string "-FIXED" attached to the end of the file name. The new file will be located in the same directory as the original. You can then use this fixed d2v file for processing. If the order parameter is set to "-1" then TFM will detect the field order from the d2v file and set the order parameter to match. TFM will also pass info from the d2v on to tdecimate to help aid duplicate detection and hybrid detection.

*NOTE: This option currently supports all d2v formats that I am aware of... which include: dvd2avi 1.76, 1.77.3 and its variants, all dvd2avidg versions, and all dgindex versions.

example => TFM(d2v="myd2v.d2v")

Default: "" (String)

ovr -

Sets the name and path to an overrides file. An overrides file allows for manual control over what matches are used for specific frames and ranges of frames, for changing the values of the field, mode, order, mthresh, and PP parameters for specific frames or ranges of frames, and for specifiying whether or not a frame should be consider combed or not combed. For more information on using an overrides file see the OVERRIDES section at the bottom of this file. You can specify an ovr file and an input file at the same time. Matches and combed frames given in the overrides file will take precedence over those in the input file.

example => TFM(ovr="myOvrFile.txt")

* NOTE: The ovr option CANNOT read output files (files made via the output option). To read output files use the input parameter.

Default: "" (String)

ovrDefault -

This is an extra option that allows you to change the default "combed" action taken when using an ovr file. By default, if you don't specify a frame as combed '+' or not combed '-' in an overrides file then TFM just processes it as usual. However, with the ovrDefault option, you can make TFM mark all frames you don't explicitly specify as either combed or not combed. Basically, its an easy way to only deinterlace a few specific frames in a file and make sure no other frames are touched or vice versa. If no ovr file is used then this setting does nothing. Possible settings:

0 - normal processing
1 - mark all frames not specified in ovr file as clean (not combed)
2 - mark all frames not specified in ovr file as combed

Default: 0 (int)

output - 

Sets the name and path for an output file. TFM will create the specified file and output all the matches and combed frame decisions used to it. Each line will contain a frame number, the match used, and the combed frame decision used. This file can then be used later for input via the input parameter (so that the matches wont have to be recalculated).

Default: "" (String)

input -

This option is meant to be used with the output option. If you have used the output option on a previous pass then to reuse that info set the input parameter to the name and path of the output file. You can specify both an input file and an ovr file at the same time. In such a case, matches in the overrides file will take precedence. The input option CANNOT read ovr files!!! To read an ovr file you must use the ovr parameter!

Default: "" (String)

outputC -

This option can be used to help identify 30fps progressive video sections by outputting the ranges over which at least "cNum" of consecutive matches were matched as 'c'. This option isn't fool proof, as slow or static film scenes can also generate long series of consecutive 'c' matches.

Default: "" (String)

cNum -

Number of consecutive 'c' matches required to output a range when using the outputC option.

Default: 15 (int)

E.) DEBUG/DISPLAY PARAMETERS:

These settings control whether debug information is output or if display information is drawn on the frames. These options can be useful when using overrides or when trying to assess matching failures.

debug -

If set to true, information will be output via OutputDebugString() during processing concerning TFM's internal operations. Info includes: match comparision statistics, match used, what matches were detected as combed (if in modes 1, 2, or 4 or PP > 0), and the values of field, mode, order, and MIC. To view this information you can use a utility called "DebugView" available for free from: 

true = enables output
false = disables output

Default: false (bool)

display -

If set to true, information will be drawn on the video frames themselves in the upper left hand corner. This info is not as detailed as the debug output and only includes the current frame number, the values of mode, field, order, and MIC the match that was used for the current frame, and whether or not the output frame was combed if PP > 0. This option can be useful when using overrides and you need to see what match is currently being used. If you need more detailed information, such as the match statistic numbers or to see if specific matches were detected as combed then you'll need to use debug output. As of v0.9.1 display will also indicate if the final output frame is combed and whether or not it was deinterlaced if PP > 0. If PP = 1 and display is enabled, then TFM will draw a white box around the area of the frame that triggered its detection as combed when a frame is found to be combed through its internal detection (this wont be done if a frame is forced to be combed through overrides).

true = enables
false = disables

Default: false (bool)

F.) OVERRIDES:

This section gives an explanation of how to specify overrides, what each type of match means, and the list of possible overrides.

BACKGROUND AND OVERVIEW:

Before we get started there is one thing that needs to be understood, and that is what in the hell the p, c, n, b, and u letters mean and where they come from ;)  Well, actually, before that we need to know what field matching actually does. I am going to assume that the reader knows what is meant by saying a stream is telecined though. Now, consider we have the following telecined stream with the bottom and top fields given as follows (1t and 1b are both assumed to come from the same progressive frame, as are 2t and 2b, etc...):

Now what TFM does, as does any field matcher, is matches the fields so that only progressive frames remain; however, it does not decimate duplicates. So, the above input stream gets transformed into one of the two following scenarios (which one depends on the field we are matching from, i.e. the value of the "field" parameter):

field = 0 (matching from bottom)

field = 1 (matching from top)

From this example, you can see why the field setting can have an impact on the matching performance... Anyways, now that we have an idea of what is happening we can move on to what p, c, n, b, and u mean. Basically, these letters just stand for what match is used. As you can see from the above example, c means that the field from the current frame and of the opposite parity to that of the field being matched from is used, p means the field of the opposite parity to the field we are matching from and from the previous frame is used, and n means the field of the opposite parity and from the next frame is used. These three matches, p, c, and n are the most common or the usual matches. In the mode parameter description these are the three matches that make up the 2-way and 3-way matching. The other matches: u and b are slightly different in that they don't use the field that is being matched from. In the mode parameter description these two matches are the 4th/5th matches. A visual explanation of all of this would probably explain things much better, so here 
are the visual representations of each match:

* In the following examples it is assumed that we are currently matching the second frame (2t/2b). Also, an "x" has been placed next to both the top field and the bottom field that will be used due to the current match.

field = 0:

field = 1:

Well, that should pretty well do it. There are a couple other points though. One, you can see from the above examples that the matches have different meanings depending on the value of the "field" parameter. That is why in all ovr files you can specify the field value with respect to which the matches in the file are made. This is so that when you specify an n, p, etc... match in an ovr file, TFM knows for sure what you mean. You don't HAVE to specify the reference field though, if you don't give one TFM simply assumes you mean to use the same value as the current internal field value. However, it is much safer to specify the value. The second point stems from the first point, and is simply that if the reference field you use for an ovr file is different from that of the internal field value then the matches will be switched accordingly so everything is the same. However, this means that if your using debug or display output the matches wont be the same as the ones in your ovr file. So for reference here are the conversions or equivalencies...

with Field = 0 ___ is the same as ___ with Field = 1 and vice versa:

p == b
c == c
n == u
b == p
u == n

SYNTAX AND USING OVERRIDES:

match codes:

* See background/overview section for visual representation of matches. For range of frames overrides you can specify multiple match codes  on the same line and the given pattern will be repeated over the length of the range.

p = match to previous field
c = match to current field
n = match to next field 
b = match to previous field (matches from opposite parity field)
u = match to next field (matches from opposite parity field)

combed frame codes:

* These can be used to manually force specific frames to be consider combed or clean. You can specify patterns of combed frame codes like you can with match codes.

+ = force frame to be considered combed
- = force frame to be considered not combed 

parameter codes:

* Parameter codes, unlike match codes, also require a value to change to when used in overrides file.

f = field
o = order
m = mode
M = mthresh
P = PP

reference field:

Since the match codes take on different meanings depending on the field parameter, you can specify a reference field for an overrides file. All the matches given in file are then taken to be with respect to the given field. If a reference field is not given then all matches are taken to be with reference to the filters internal field value.

To specify the reference field you must put one of the following two lines as the VERY FIRST active line in the file:

field = top 
field = bottom

The first one would set the reference field to top (1) and the second would set it to bottom (0). For an example see the included example ovr file included in the filter zip file.

override syntax:

single frame => frame_number code(s) [change_value]        [] = only required for parameter codes
frame range => start_frame,end_frame code(s) [change_value]

examples:

0,100 pcpcbuu will repeat the given pattern "pcpcbuu" of matches starting at frame 0 and ending at frame 100

other things to know/remember:

1.) Ranges are inclusive.

2.) For parameter codes, the changed value is always set back to what it was originally set to after the override goes out of the specified range. (i.e. if you specify a field override for frame 600 to 700 after frame 700 field is set back to its original value automatically, you don't need to set it back in the overrides file! The original value is what it is set to on load (i.e. either the default or what you set it to in your avisynth script).

3.) Frame numbers must be within range for the file.

4.) Frame numbers for match codes must be ascending, and frame numbers for the same parameter codes must not overlap!!! In other words, don't make multiple overrides for the same parameter over the same range of frames.

5.) The spacing is important! (no spaces between the comma and frame numbers, one space between the end frame number and the specifier, one space between the specifier and the value to change to if it has one). Just look at the examples for clarification.

6.) Only match codes and combed frame codes can be used in pattern specifications. Pattern specifications can be up to 50 match codes maximum.

7.) To comment out a line (have TFM ignore it) in an ovr file, you can either put a '#' symbol or ';' at the beginning of the line.

8.) An example ovr file w/ comments is included in the filter zip file for reference.

9.) Setting the second frame in a range to 0 means the same as setting it to the last frame of the clip.

G.) Changelist

06/25/2005 v0.9.8.1
+ Added support for new dgindex d2v's (d2vformat v10)

06/06/2005 v0.9.8
+ Added a new match mode (mode 2, old mode 2 is now mode 3, 3 is 4, and so on)
+ Added support for new dgindex d2v's (d2vformat v9)

04/20/2005 v0.9.7.3
+ Debug and display options now output a value called MIC (the value that is computed and compared against MI to check if a frame is combed) to make tweaking the MI parameter easier. The MIC value is computed during combed frame detection and is the single highest value computed for any one of the blockx by blocky sized blocks. Previously, combed frame detection exited immediately upon finding the first block with an MIC value greater then MI... it now scans all blocks and returns the highest. Thus, the white box that is drawn when display=true, PP=1, and a frame is detected as combed now corresponds to the block with the highest MIC value and not the first  to be found.
- Changed default MI value to 85 (also changed to 85 in IsCombedTIVTC)
- Fixed the white box drawn on combed frames when PP=1 sometimes being drawn in the incorrect spot when mode was set to 1, 2, or 4

03/13/2005 v0.9.7.2
+ Post-processing now only links luma->chroma on the motion map, instead of doing full luma->chroma, chroma->luma, and chroma->chroma linking.

03/10/2005 v0.9.7.1
- Fixed a crash that would occur if the d2v option was enabled and a trim() or some other filter that altered the number of frames was used prior to tfm().

02/19/2005 v0.9.7
+ Added support for new dgindex project file versions 07 and 08 (dgindex v1.2) to the d2v option.
- Fixed the d2v method of fixing illegal transitions.
+ TFM now uses the rff flags from the d2v to help tdecimate identify duplicates and 24p film.
- Changed default MI value to 100.
- Fixed the readme file indicating chroma=true, when it actually defaults to false.

01/08/2005 v0.9.6.3
- Fixed SetCacheHints being called incorrectly and always defaulting to CACHE_ALL.

01/02/2005 v0.9.6.2
- Changed the d2v method of fixing illegal transitions to one that doesn't alter the total # of fields in the clip or the # of fields from each of the original frames in the stream.
- Changed default value for the chroma option to false.
+ Some other optimizations and internal changes.

12/22/2004 v0.9.6.1
- Fixed passing AsClip() a NULL argument during TFM and TFMPP object creation if no clip was given for clip2. This would cause assertion failure when compiled in debug mode.

12/19/2004 v0.9.6
- Fixed not invoking a cache between TFM and TFMPP when PP > 4.
+ Some internal additions to allow TDecimate to tell the difference between c matches and deinterlaced frames.
+ Added debug output to d2v checking... when debug=true and a d2v file is specified it will report the type of d2v detected, the detected field order from the d2v if order = -1,
and report the errors if any are found.

11/30/2004 v0.9.4

+ significant speed up (30-50%) to modes 1, 2, and 4 with PP > 0, due to some missing logic the output frame would needlessly get checked twice for combing.

11/28/2004 v0.9.3.1
- Fixed crashing when mode = 0 or mode = 3 and PP was greater then 0, the conditions for an array allocation were not adjusted as needed when the PP option was added in v0.9.1.

11/28/2004 v0.9.3
+ Added crc checking to input/output files. Whenever an output file is created, TFM creates a crc32 value based off the first 15 frames in the current clip. When the file is loaded via the "input" option the crc value from the file is checked against the current clips crc value.

11/27/2004 v0.9.2
- Fixed "slow" and "mChroma" parameters defaulting to false instead of true.
- Fixed display option not writing any part of a line if the total line length exceeded the width of the frame instead of writing what would fit.
- Changed default cthresh value to 10.
- Changed default PP value to 6 (motion-adaptive cubic interpolation).
+ Added d2v and ovrDefault options.
+ Added IsCombedIVTC() conditional function.
+ Added cubic deinterlacing method.
+ Re-did input/output parameters. Files created via the output option can no longer be read via the ovr option! You must use the input parameter to read files created via the output parameter!.
+ Output option no longer requires linear access.

11/21/2004 v0.9.1
- Changed default mode value to 1.
- Changed default MI value to 128.
- Allow blockx and blocky sizes up to 2048.
- Reordered the parameter syntax slightly.
+ Added exclusion band (y0/y1).
+ Added post-processing (PP parameter) and related stuff.
- Added clip2 and mthresh parameters.
+ Added hinting in preparation for tdecimate.

11/11/2004 v0.9.0
- initial release

$Date: 2005/07/10 16:11:01 $