MAME Movie Maker released

[Howard_Casto]

budda... if you are interested... i've been playing around and I think i have a solution for the several fighters that don't get their loop detected properly.

You need to add the option to do a sample check after a possible loop is found.  Try a simple rgb point sample in each corner and in the center.... if the colors all match up... keep going.  This way the loop only detects loops with actual frames, not solid color fields and screen wipes. 

So the loop would start and end with the first frame of the title screen, not the blank wipe before it. 

[donnyj]

Thanks buddabing for helping me fix it!

Thanks everyone else for telling me you compiled from scratch too, was worried I was just making my life hard!!!

Now I'll try making cpmaker, just to punish myself some more! 

[Buddabing]

budda... if you are interested... i've been playing around and I think i have a solution for the several fighters that don't get their loop detected properly.

You need to add the option to do a sample check after a possible loop is found.

[Howard_Casto]

budda... if you are interested... i've been playing around and I think i have a solution for the several fighters that don't get their loop detected properly.

You need to add the option to do a sample check after a possible loop is found.

[Buddabing]

Howard,

I coded up your idea except I am sampling the whole image and taking the maximum and minimum r,g,and b values over the whole thing. The frame is skipped if the max r minus the min r is less than the -moviethreshold parameter, and the max g minus min g, and the max b minus the min b.

I need more specific information on what games are having a problem. "Allfighters made by x,y,z" means nothing to me since I don't play fighters. I need rom names, not game names.

You mentioned 64street and arabfgt earlier. 64street has a static picture on it with changing text below the picture. A loop is detected when the text is blanked out and just the picture is showing. The new method of blank frame detection does not detect this.

On arabfgt, a loop is detected at a screen blank also, but this time the screen is not completely blanked, it has the "insert coin" text at the top and "sega 1991" at the bottom.

A black detection that excluded the top and the bottom of the screen would help arabfgt but not 64street, and may break other games.

I agree with you that having the game's title screen at the beginning of the video is important. I just don't think the solid color blanking is as common as you think.

But, give me a sample of a game with this solid color blanking, and I'll see if this changed blank frame detection will help.

In fact, you can try it yourself. I've uploaded the test executable here.

[2600]

Howard,
I believe most/all of these can be solved by changing the minmovielength field.  I put in 5200 and ran in on the kof series(you mentioned them earlier), aof, and aof2. I beleive all made the movies fine except for aof and kof94.  Those would of probably been made fine if I picked 5100 instead or changed one other field.

Now you may argue 7 out of 9 aren't good enough odds, but realize this was one number I picked (on purpose not out of my @) and made the the movies in what an hour since your last post.  If I tuned it I probably could of gotten them all.  And because I logged them to a file, I can post the start and stop frames to Silver so everyone can make them without any effort.

You impiled that you aren't sure we know what an attract mode is.  Can you try the above and verify that the movies were attract modes would be captured properly?

Edit: I started to look at the movies and the first one kof2000 wasn't completely done (it only did one of the fight scenes and the characters change on each loop so a proper loop would get all of the fight scenes), but it still can be done properly if you change the minmovielength and maxmovielength field.  I imagine the rest need a bit more modification as well.

[Buddabing]

In case anyone missed it I've cleaned up and put the new ssf.txt file on my website in the format:

romname,startframe,endframe

there are no endframes yet (obviously) as this was only used to identify end-of-bios points.

Its worth checking these points, and it is an old file so there is a (slim) chance that some of the romnames may have changed slightly...

Cleaned up file: http://www.silverfoxy.plus.com/ssf-csv.txt

Original file: http://www.silverfoxy.plus.com/ssf.txt

I'll happily host updates as people work through the games.

Silver:

Please post a link to the up-to-date modified csv. I'll repost it to the first page of this thread.

[Silver]

Hi,

At the moment I've not had anyone send in contributions, and due to my main pc being down-and-out I can't add my own additions so far - sorry. I've got a MySQL database online waiting, but can't add a web interface till the pc is rescued from the dead. Hopefully not too long, but I've got a mad week at work or so.

If anyone wants to email me there progress so far (either a .csv file or even just a collection of successful .vdb files) I'll throw it into a file and post the link here - I would be able to keep that updated every day no problem until the online db is running.

email:  silver@silverfoxy.plus.com

or just PM me.

[Howard_Casto]

Howard,

I coded up your idea except I am sampling the whole image and taking the maximum and minimum r,g,and b values over the whole thing. The frame is skipped if the max r minus the min r is less than the -moviethreshold parameter, and the max g minus min g, and the max b minus the min b.

I need more specific information on what games are having a problem. "Allfighters made by x,y,z" means nothing to me since I don't play fighters. I need rom names, not game names.

You mentioned 64street and arabfgt earlier. 64street has a static picture on it with changing text below the picture. A loop is detected when the text is blanked out and just the picture is showing. The new method of blank frame detection does not detect this.

On arabfgt, a loop is detected at a screen blank also, but this time the screen is not completely blanked, it has the "insert coin" text at the top and "sega 1991" at the bottom.

A black detection that excluded the top and the bottom of the screen would help arabfgt but not 64street, and may break other games.

I agree with you that having the game's title screen at the beginning of the video is important. I just don't think the solid color blanking is as common as you think.

But, give me a sample of a game with this solid color blanking, and I'll see if this changed blank frame detection will help.

In fact, you can try it yourself. I've uploaded the test executable here.

The zip file you've linked to isn't downloading properly... it gets corrupted.  I'll be glad to try it though. 

It's an extremely common problem, but since you don't play fighters, you'll never see it.    It happens on virtually all snk and virually all capcom fighters.... that's 600 or so odd parents, which is a noticable chunk of mame. 

as i said.... snk fighters are the most effected.....

take a look at aof-aof3 and the kof games  I think some of the metal slug games might also suffer from it, but I can't sware to it. 

Besides that... your loop detection is just "too good".  Even if there isn't an apparent transition, many games have single frames of either a blank screen or a colored screen several times durring the attract mode.  If it happens enough times, the loop detection will fail. 

Let me know what I can do to help.


About 2600's comments:

I'm sure that if I adjusted the minmovie length to an insainely high number it might fix some of the games, but at the cost of "breaking" many of the old vector games as well as any game with a really short attract mode.   

More importantly, you hit the nail right on the head.... 7 out of 9 isn't good odds.  For me at least, 8 out of 9 isn't good odds either.  This project is great, but it's only useful if the ratio is something like 98%, not 80. 

Remember... the whole idea is to be able to use a giant batch file and have mmm automatically make the vids.  If we have to be tweaking numbers for each game then it's easier to just use )p( or my capture rigs on those games. 

[Buddabing]

Howard,

I coded up your idea except I am sampling the whole image and taking the maximum and minimum r,g,and b values over the whole thing. The frame is skipped if the max r minus the min r is less than the -moviethreshold parameter, and the max g minus min g, and the max b minus the min b.

I need more specific information on what games are having a problem. "Allfighters made by x,y,z" means nothing to me since I don't play fighters. I need rom names, not game names.

You mentioned 64street and arabfgt earlier. 64street has a static picture on it with changing text below the picture. A loop is detected when the text is blanked out and just the picture is showing. The new method of blank frame detection does not detect this.

On arabfgt, a loop is detected at a screen blank also, but this time the screen is not completely blanked, it has the "insert coin" text at the top and "sega 1991" at the bottom.

A black detection that excluded the top and the bottom of the screen would help arabfgt but not 64street, and may break other games.

I agree with you that having the game's title screen at the beginning of the video is important. I just don't think the solid color blanking is as common as you think.

But, give me a sample of a game with this solid color blanking, and I'll see if this changed blank frame detection will help.

In fact, you can try it yourself. I've uploaded the test executable here.

The zip file you've linked to isn't downloading properly... it gets corrupted.  I'll be glad to try it though. 

It's an extremely common problem, but since you don't play fighters, you'll never see it.    It happens on virtually all snk and virually all capcom fighters.... that's 600 or so odd parents, which is a noticable chunk of mame. 

as i said.... snk fighters are the most effected.....

take a look at aof-aof3 and the kof games  I think some of the metal slug games might also suffer from it, but I can't sware to it. 

Besides that... your loop detection is just "too good".  Even if there isn't an apparent transition, many games have single frames of either a blank screen or a colored screen several times durring the attract mode.  If it happens enough times, the loop detection will fail. 

Let me know what I can do to help.


About 2600's comments:

I'm sure that if I adjusted the minmovie length to an insainely high number it might fix some of the games, but at the cost of "breaking" many of the old vector games as well as any game with a really short attract mode.   

More importantly, you hit the nail right on the head.... 7 out of 9 isn't good odds.  For me at least, 8 out of 9 isn't good odds either.  This project is great, but it's only useful if the ratio is something like 98%, not 80. 

Remember... the whole idea is to be able to use a giant batch file and have mmm automatically make the vids.  If we have to be tweaking numbers for each game then it's easier to just use )p( or my capture rigs on those games. 

Howard,

I must have forgotten to set binary mode when uploading that zip. Sorry about that.....

I had an idea which I want to bounce off of you and anyone else reading this thread. Currently, MMM detects the end of a loop by matching exact frames. So, if frame number 100 and 500 are identical, a loop is detected. But, as we have seen, there are many games which give false matches. Now suppose we introduce a new parameter, -moviematchlength. If this parameter is one, it does the matching the exact same way it does now. Let's say we set the parameter is 10. So, frame 100 and 500 can match, but frame 101 and 501, 102 and 502, etc. have to match all the way up to 109 and 509. Considering the case of 64street, the loop is detected when the text is cleared below the picture. If the length of the match was changed, the number of matched frames would be below the -moviematchlength.

The more I think about it, the more I think it will work and solve a lot of the problems.

LMK your thoughts.

[Silver]

Hey I really like that idea.... I don't think you'll even need a very high value to take advantage.

The only place I can see it fall down is if some games have a loop-within-a-loop...  ie if the attract mode is loop A then loop B then loop A then loop C etc.... but I can't think there are many of those.. Good idea!

[Buddabing]

Success!

I changed the code tonight for this idea of mine and it works. I tested kof94, 64street, and arabfgt. All these games had places which would have loops detected but now passed through.

I used a value of -moviematchlength equal to 10. 64street had a false match of length 7, and another game had one of length 5.

You don't want to use too large a value of -moviematchlength. Zookeeper was affected but not too badly. Other games may be affected adversely. For now I will set the default value to 10.

I will upload a new binary and source tomorrow.

[donnyj]

I had an idea which I want to bounce off of you and anyone else reading this thread. Currently, MMM detects the end of a loop by matching exact frames. So, if frame number 100 and 500 are identical, a loop is detected. But, as we have seen, there are many games which give false matches. Now suppose we introduce a new parameter, -moviematchlength. If this parameter is one, it does the matching the exact same way it does now. Let's say we set the parameter is 10. So, frame 100 and 500 can match, but frame 101 and 501, 102 and 502, etc. have to match all the way up to 109 and 509. Considering the case of 64street, the loop is detected when the text is cleared below the picture. If the length of the match was changed, the number of matched frames would be below the -moviematchlength.

The more I think about it, the more I think it will work and solve a lot of the problems.

LMK your thoughts.

I like this idea a lot.  My thoughts are having some kind of "variability" for the detection.  Why?  Because as CPU speeds are changed/corrected, and emulation improved/deprecates, the attract mode won't get recorded wrong.  That is why I stopped doing start/stop frames.  The numbers may work for one person on one build, but will they work on another machine with another build?  How about using ASM cores and different optimizations...

BUT, having a detection scheme (that works) should be a lot more robust to any code / optimization changes / quirks.

Thanks for all the hard work buddabing and others!!!  I believe this is a worthwhile project!

[donnyj]

Related, but unreleated:  Anyone know how to do a "diff" and have it act on all files in a directory and subdirectories?

I manually added buddabing's changes to the mame 91u2 source and I would MUCH prefer to take future versions of MMM and diff them with the version before it and manually make the changes myself...  Probably not making sense =[

TIA

[Buddabing]

Related, but unreleated:

[Buddabing]

I've uploaded a new version with the -moviematchlength parameter implemented. The default value of the option is one to match the existing loop detection. I've found a value of 10 to work nicely. The value can range from 1 to 99.

Also, I've uploaded a corrected version of Howard's test black detection which detects all monochromatic images, not just black. The test black detection, testing all pixels instead of a subsample, runs considerably slower on Zookeeper, 17.41 fps versus 26.38 for the standard black detection.

I'm thinking maybe I can get by without any black detection at all it the -moviematchlength is sufficiently high. But, that wouldn't work for games with long stretches of black (greater than 10 frames) within their demo loops.

EDIT:
I've added a new parameter, called -movieskipblack. It's in the test executable only. A value of 0 for -movieskipblack does not skip black frames when calculating loops. A value of 1 for -movieskipblack will ignore frames which have at least one pixel's red,green,or blue value above -moviethreshold. A value of 2 for -movieskipblack will calculate the maximum and minimum red, green, and blue value for each frame and will ignore frames which have the maximum red minus the minimum red less than -moviethreshold, and likewise for green and blue. This "ignoring" is for the purposes of loop detection only.

I added this parameter because I want to test getting rid of the black detection altogether. It runs faster that way.

[papaschtroumpf]

when you say you uploaded a new version, is it just the binary or the sources too?
I assume that if there is new source, the safest is to erase the whole source tree and re-apply the changes to mame 0.91u1?

[Buddabing]

when you say you uploaded a new version, is it just the binary or the sources too?
I assume that if there is new source, the safest is to erase the whole source tree and re-apply the changes to mame 0.91u1?

That is the safest thing to do. Typically, all I do is extract the new source over my existing source, then go to the obj/mame folder and del *.o and deltree windows. This will force a recompile, but it won't recompile the cpu cores and drivers, which saves a good bit of time.

When I upload a new version it is the binary and sources. For the test black detection I wrote for Howard it was just the binary.

[Howard_Casto]

Hey I really like that idea.... I don't think you'll even need a very high value to take advantage.

The only place I can see it fall down is if some games have a loop-within-a-loop...

[Buddabing]

I had an idea which I want to bounce off of you and anyone else reading this thread. Currently, MMM detects the end of a loop by matching exact frames. So, if frame number 100 and 500 are identical, a loop is detected. But, as we have seen, there are many games which give false matches. Now suppose we introduce a new parameter, -moviematchlength. If this parameter is one, it does the matching the exact same way it does now. Let's say we set the parameter is 10. So, frame 100 and 500 can match, but frame 101 and 501, 102 and 502, etc. have to match all the way up to 109 and 509. Considering the case of 64street, the loop is detected when the text is cleared below the picture. If the length of the match was changed, the number of matched frames would be below the -moviematchlength.

The more I think about it, the more I think it will work and solve a lot of the problems.

LMK your thoughts.

I like this idea a lot.

[2600]

I've played around with -moviematchlength.  Works really well, especially for the games that were causing problems aqua jack, kof series, etc.

I don't know if we will find one setting that works with everything, but your findings about varying the movielength per year are what I was going to suggest.  Although, one other parameter should be adjusted as well -minmovielength.  The reason I suggest this is -minmovielength 800 didn't work for kof96 because it has a really long section in the middle where it displays the name and has the flame/clouds at the bottom.  For games above 1990, I'd say use a -minmovielength of around 2000.

There is one more idea for games like Howard was talking about TMNT or the MK series.  Here are the start and stop for MK series so you can see the entire attract mode.
mk3,178,23553
mk2,60,18251
umk3,2369,22852

I'm sure you can think of a better idea, but here are two.
1. Add a -movieloop, this would require subsequent loops have to match the first.  So the loop detection would keep going until a loop matches the first, but only makes a video for the one.
2.  A slight modification to the except with a -moviesamples where instead of matching the loop entirely, you match every 50 samples.  So loop 2 frame 0,50,100 have to match loop 1's respective frames.

The only problem with this is when I was trying kof2000, it appeared that the fight scene was generated randomly.  (Howard is that your view on it)


Howard,
You were right about the odds and to accept lower odds, but I had posted too fast.  Maybe I should of worded that it is possble, where I believe you were saying it isn't.  Or Maybe I've been reading too many of your posts in getting the user's to learn how to fish.    In any case, I'm not sure if one bat file will ever be possible, but the odds are getting higher.

[Silver]

Hey good progress - this new addition is looking good!

Here are the start and stop for MK series so you can see the entire attract mode.
mk3,178,23553
mk2,60,18251
umk3,2369,22852

Heyhey! First additions to the collection. Thanks.. 

[2600]

Heyhey! First additions to the collection. Thanks.. 

Yeah, sorry about that.  I've got a list, but haven't verified them.  Don't want to post numbers that I'm not positive about.

[Silver]

No problem...

The way progress is ticking along here - kudos to all for the ideas and BuddaBing for implementing them so quick - it won't be too long till we can run 1 long batch and get a very high success rate to start with. Then its just a case of spotting any flaws....

[Buddabing]

I've played around with -moviematchlength.  Works really well, especially for the games that were causing problems aqua jack, kof series, etc.

I don't know if we will find one setting that works with everything, but your findings about varying the movielength per year are what I was going to suggest.  Although, one other parameter should be adjusted as well -minmovielength.  The reason I suggest this is -minmovielength 800 didn't work for kof96 because it has a really long section in the middle where it displays the name and has the flame/clouds at the bottom.  For games above 1990, I'd say use a -minmovielength of around 2000.

There is one more idea for games like Howard was talking about TMNT or the MK series.  Here are the start and stop for MK series so you can see the entire attract mode.
mk3,178,23553
mk2,60,18251
umk3,2369,22852

I'm sure you can think of a better idea, but here are two.
1. Add a -movieloop, this would require subsequent loops have to match the first.  So the loop detection would keep going until a loop matches the first, but only makes a video for the one.
2.  A slight modification to the except with a -moviesamples where instead of matching the loop entirely, you match every 50 samples.  So loop 2 frame 0,50,100 have to match loop 1's respective frames.

The only problem with this is when I was trying kof2000, it appeared that the fight scene was generated randomly.  (Howard is that your view on it)


Howard,
You were right about the odds and to accept lower odds, but I had posted too fast.  Maybe I should of worded that it is possble, where I believe you were saying it isn't.  Or Maybe I've been reading too many of your posts in getting the user's to learn how to fish.    In any case, I'm not sure if one bat file will ever be possible, but the odds are getting higher.

I'll make -moviematchlength 0 equal to "let the application decide" as well as -minmovielength (it really should be -movieminlength) 0. You can be the arbiter of what the application decides.

Your other suggestions are getting into the realm of diminishing returns. It would probably be best to treat multiloop games as exceptions and just get the start and end frame manually like you did for mk2.

The loop detection doesn't currently keep track of where the match occurs, only that a match happened n times in a row. So maybe an improvement is possible there.

[donnyj]

Like I originally said quite awhile ago (got a lucky guess), I believe the best system will end up being "general generic rules", with manual "hints"

For example

Game1, 3 loops
Game2, automatic
Game3, end on frame 25000

and the general generic rules could vary internally to handle situations like: fighting game between 1990-1995

Quite complicated but I believe quite workable (of course I don't do the programming so what do I know?)

anywho, a solid generic rules system with "hints" for the games that don't work well with those...

I am excited about the changes that are being made and the diligence of the people working on it (especially budabing and silver)

[Buddabing]

Like I originally said quite awhile ago (got a lucky guess), I believe the best system will end up being "general generic rules", with manual "hints"

For example

Game1, 3 loops
Game2, automatic
Game3, end on frame 25000

and the general generic rules could vary internally to handle situations like: fighting game between 1990-1995

Quite complicated but I believe quite workable (of course I don't do the programming so what do I know?)

anywho, a solid generic rules system with "hints" for the games that don't work well with those...

I am excited about the changes that are being made and the diligence of the people working on it (especially budabing and silver)

I'm mulling over how to implement this "rules" system.

What would be cool is a genrules program which takes the rom name as a parameter and spits out a .ini file with the correct movie parameters.

The genrules program would operate with a SQL engine:

-- defaults
set minmovielength='800';
set maxmovielength='10000';
set moviematchlength='1';

set minmovielength = '2000' where year > '1989' and catver_category like '%Fighting%";
set maxmovielength = '25000' where romname= 'mk3';
set moviematchlength='10' where ...... etc.

The genrules program would be called if -minmovielength or -moviematchlength was equal to zero.

It would add a constant amount of time, I figure a couple of seconds or so, to the creation of each movie, but it would be pretty fast.

I would just adapt the source from my ListGen program which already has the SQL stuff.

The question remaining to me is: is it worth it? Does the latest loop detection method meet Howard's criteria? If not, will this improve it?

Please post your thoughts.

[donnyj]

VirtualDub Build 23174 (1.6.4, experimental): [February 12, 2005] was released

Among A LOT of other changes are:

    * PNG decoder failed if the compressed data contained a stored chunk.
    * PNG decoder crashed if an error occurred during decoding.

[papaschtroumpf]

meaning what? use an older version for MMM?

[donnyj]

meaning what? use an older version for MMM?

Meaning the last version that we used for MMM choked on uncompressed PNG files, and the problem is now fixed.....

So, if you want to try uncompressed PNG files, upgrade 

[Buddabing]

VirtualDub Build 23174 (1.6.4, experimental): [February 12, 2005] was released

Among A LOT of other changes are:

    * PNG decoder failed if the compressed data contained a stored chunk.
    * PNG decoder crashed if an error occurred during decoding.

This pretty much makes the Crashtest mng files redundant. Using MMM and this new VirtualDub we can create lossless AVIs with as good or better resolution.

[Howard_Casto]

which is great since mngs are the most un-supported file format since ogg vorbis.   

[donnyj]

Really stupid question:  Is the "-resolution" command to prevent games that change resolution to change or something else?  Because when I run without that command, I get the same output (on the few games that I checked)

Also, just to be a pain in the butt  When's 0.92 gonna be ready?    Just kidding, whenever it's ready is find, just had to throw that in

[Buddabing]

Really stupid question:

[papaschtroumpf]

there we go again! 

[donnyj]

Does anyone have any feedback on my idea on the rules generator?

The reason that I haven't commented is that I don't want to be responsible for so much effort    I believe that having a rules-based system will allow creating a settings file that will allow anyone, anywhere to run moviemaker and get the desired results.  As they change drivers, optimizations, cpu cores, fix timings, etc. I think that frame numbers could change...  thus unless someone is willing to manually check each change on every movie, a rules-based system should help....

I have a separate request.  Could you please make an optional switch that tells MMM to continue running and saving frames until  the maxmovielength is reached.  I currently run MMM into a different folder for each gamename....  I can either run MMM twice, the first time with loop detection enabled, then delete the images and re-run with loop detection disabled...  to verify that MMM did indeed find the proper end of the loop(s).   With the "debugging" switch on, I could run MMM once, verify the detected frames as correct, and continue....

The reason I asked about the resolution switch is that some of the artwork (especially for the vector games) doesn't show unless artwork cropping is disabled, and when I did that, without specifying a resolution, it "seemed" to create the proper resolution for the game, and still showed the artwork...

[Howard_Casto]

Just my 2 cents.... I think anything resembling a "script" jsut to get a movie to record is silly.  If it's that much trouble then do it manually. 

I tested .92 and I was pleasently suprised by the results.  The problem games seem to work now either either the blank option set to skip solid colors or the matchlength set to 10.  Both seem to work pretty well, but as you expected the match length option is much faster.  I obviously haven't had a chance to try em all. But out of the 20 fighters I did today, only C&d (dino.zip) and Bloody roar 2 screwed up.  And C&D is almost certain to screw up anyway due to it's strange attract mode. (It shows the title screen 2 to 3 times before it actually gets to the gameplay sample)

I've noticed a few oddities though and I'm wondering if it's the mmm or the games themselves.  Bloody Roar 2 seems to freeze halfway through it's attract mode.  It uses an video file for it's attract mode so that may be the culprit. 

One other thing and I think this might be a game thing.  Some games seem to get the frist few seconds of their attract mode cut off, yet they are present at the end of the loop.  Does this have something to do with the code you are using?  I honestly doubt it, but I thought I would ask. 

I think at this point the loop detection is about as robust as it's gonna get.  I would suggest that we all start playing with settings and report back with what works best.... then budda can figure out a good set of default settings, use those as the defaults in his code, and the setup would become farily turn key. 

Great job, all in all.  I think this should just about do it.   

[Lilwolf]

I would love if when you download it, a script of all known good settings per game...  So you run it, go to bed... wake up with all known good loops and start/end for those who dont...

Of course... this would take a TON to upkeep... And the start/end would have arguments (but it doesn't wait long enough for cammies 3rd intro sequence)...

but it still would be cool to get a list of all known working settings for each game... and just start off with a decent chunk of known working.

[Buddabing]

I can get a bit manic-depressive on some of these projects, so adding more features isn't a problem when I'm in manic mode.

Howard: I unfortunately do not have Bloody Roar 2. Sounds like an interesting game, though. The loop detection works as designed on C&D, though. This is a game that would benefit from a "multi-loop" option. And what about your 98%? A nice set of rules should accomplish it.

When a loop is detected, the last -moviematchlength frames are subtracted out from the start and the end of the loop, to account for the match length. As an example, suppose a loop is found between frames 123 and 456. The program will start detecting a loop at frame 456 and will continue detecting until frame 465. At that point, the corresponding frame to frame 465 is frame 132. To correct the loop, the program subtracts -moviematchlength and adds 1 to both start and end frame. So 132 becomes 132-10+1=123 and 465 becomes 465-10+1=456. I'll look at the code and I'll make sure the .vdb file is being generated correctly taking the correction into account.

Donny: I've created a -movieloops 3 option. It'll detect a loop but will continue capturing frames. The resulting movie, however, will not have the detected loop, it will be the entire span of frames. Do you want to have the original loop in the movie? Also, the -movieloops option is set by the batch.bat, so you will have to edit your makemovies.bat and set the -movieloops there.

Lilwolf: The mechanism is already there. Silver is compiling a list of exceptions. This list can be read by my ListGen program which generates a big batch file for MAME to generate movies from which have the start and end frame in them. You can even filter out the movies which have exceptions.

When I ran a batch of 2000+ games last month, it took a solid week.

There's a new binary which has the -movieloops 3 option.

[Silver]

Hi,

Great Progress with 0.92 build. Think I will put this one into gear and get my pc churning out videos, and building up the frame info. (Now that my pc is back up and running, after replacing every cap on the board......)

@Buddabing

It appears Vdub 1.6.4 has a slightly different Script format. For me it errors out with an unhelpful "" error AFTER making the video. The video is fine, but it needs manual intervention to continue. The culprit appears to be the:
"VirtualDub.video.SetDepth(24,24);" line in the cdb file. Removing this works fine - it appears this is no longer recognized in 1.6.4. It would be really helpful if you could remove it from the mame build.....

Also, when supplying the start and end frame inputs, does it still run the dectection routine? I would have expected to be slightly faster due to skipping the detection, but maybe it has no impact.