MAME Movie Maker released

[Buddabing]

yes I agree that

[donnyj]

If I understand what papaschtroumpf is saying, he wants to double the resolution of the game itself.  Doubling the resolution of the PNGs is only going to display at the same quality as before.

papaschtroumpf: I haven't found a game yet that the COMPLETE loop extends beyond 20,000 but I am sure a few do.... I believe the best solution currently available would be to extend the minimum frame number...

Buddabing:  how does MMM work?  If I knew a game had the same title sequence, but a different between sequence pattern:

T = Title Screen Loop

T {demo with fighter plane1} T {demo with fighter plane 2} T {demo with fighter plane 3} T {demo with fighter plane 1}....

What would be the best way to detect the end of the loop after plane 3, but before it restarts the loop for 1 again?   Skip to near the end of plane 3 with minmoviestart or minmovielength or something else?

Donny

[donnyj]

Problems please confirm or rebuke these issues...

I ran the # movies that:

select * from gamedata where screen not like 'vector' and status like 'good' and gamelist_working like "Yes";

format:
gamename : Correct/Wrong : Frame Start ( Round(29.97*SetRange/1000) ) [, correct is <frame#>]
1941 : Correct : 860 (28695)
1942 : Wrong : Reported 640 (21354), correct is 5
1943 : Wrong : Reported 5010 (167167), correct is 551
1943kai : Correct : 551 (18385)
1945kiii : Wrong : Reported 52 (1735), correct is 45
19xx : Wrong : Reported 4082 (136202), correct is 1961
2020bb : Wrong : Reported 231 (7707), correct is 222
3countb : Wrong : Reported 231 (7707), correct is 222
3wonders : Correct : 667 (22255)
4dwarrior : Correct : 17 (567)
4enraya : Correct : 204 (6806)
4in1boot : Wrong : Reported 133 (4437), correct is 110
4psimasy : Wrong : Reported 416 (13880), correct is 366
64street : Wrong : Reported 195 (6506), correct is 9
720 : Wrong : Reported 148 (4938), correct is 111
88games : Wrong : Reported 504 (16816), correct is 248
8ballact : Wrong : Reported 264 (8808), correct is 24

Why are we using Setrange () at all?

[Silver]

Well I'm not sure about the correct/incorrect values.

Setrange/Addrange:

Its possible these have changed in recent Vdub updates, but all the docs for scripting only go up to Vdub 1.5.10, and AddRange does not even exist in those.

However, if I recall correctly (cant test right now) we need both so that we can edit the audio to match.

ie we include ALL the pngs in the AddRange when we load vdub, so that we have synced audio/video. Then we tell Vdub only to process what we want with the Setrange command, and it trims BOTH the audio and video.

If we only used Addrange, say starting from frame 500, we would then have to externally trim the wav file correctly. In fact this was how we did it in the original version of MMM by using AviSynth to sync and trim the the video and just feed it into vdub for compressing (Vdub did not support pngs then).

I've emailed about this before (to the Vdub author) and can only hope he allows us to use a frame number in Setrange so we can avoid this silly x29.97 fudge.

[Buddabing]

If I understand what papaschtroumpf is saying, he wants to double the resolution of the game itself.

[2600]

Not sure if this one is for Silver or Buddabing, but I was doing some thinking and testing and I noticed the neogeo games are going about 10 frames too far.  It may not just be neogeo, but it was easy to tell because the neogeo bios started up again at the end.  Using latest 94 release and virtualdub 1.64.

I'm not sure if this relates to my moviematchlength, if the checksums don't match, or what.

[Buddabing]

Not sure if this one is for Silver or Buddabing, but I was doing some thinking and testing and I noticed the neogeo games are going about 10 frames too far.

[2600]

I did a length of 50 and it was definitely longer than 10 frames, more like 50 or 60(i-fun viewer is great). Also, can you explain this.  Here is the output of the last test. Why is Mames FPS Total # frames, less than the other counts?

50-frame match detected between frame 334 and 10212
movie will run between frames 285 and 10163
Average FPS: 33.205449 (10152 frames)


I suppose another possibility is the 29.97 thing, but it does seem to have some correlation with moviematchlength.


BTW, MMM didn't accept 100 as a moviematchlength argument.

[Buddabing]

I did a length of 50 and it was definitely longer than 10 frames, more like 50 or 60(i-fun viewer is great). Also, can you explain this.

[2600]

Buddabing,
Yeah, 99 is plenty.

I was mainly testing with 3countb I tried it with another neogeo game, aof I believe.  I didn't try it with a non-neogeo game, because I was testing something with the neogeo games.  Let me know if you want me to look at something else though.

[Buddabing]

It might be smart to run each game for a few frames before generating the movies so that the nvram is generated. There was a method for doing this previously in this thread.

anyone knows what the option is off hand? I haven't founf it in the thread yet or in the windows.txt file from MAME.
EDIT: the -ftr option is what I was looking for, finally found it in WINDOWS.TXT
I added "mame %1 -ftr 120 -window -resolution %2" before the call to mame that actually generate the movie. Adds a few seconds to the process but it's not that big a deal. I could even check for the presence of an .nv file in the nvram directory to avoid this step when the game has been ran.

I've added the -ftr 100 nvram generator and a check for %1.nv into the batch.bat on the first page. Good suggestion!

[donnyj]

I'll look at your frame number issue.

Any luck?

[donnyj]

Buddamame:  Did you fix the aspect ratio issues or did mamedev?  Because now all my screenshots are MUCH better?

[Buddabing]

Buddamame:

[2600]

Buddabing,
I may have found the bug and that I think will start you in the right direction.  If you run it with the a log on, you'll notice that you check the MD5 sum on every other frame.  So for a 10 frame match you have to actually run 20 frames because you are checking every other frame.  You identify the correct end frame when you start matching, but somewhere after I assume is a miscalculation based on the fact that you are on frame 20 instead of 10.

Hopefully this is descriptive enough, I'm in a hurry.

[Buddabing]

Buddabing,
I may have found the bug and that I think will start you in the right direction.

[2600]

Using Mame .94 and testing with 3countb.   I am using what you suggest for the mame.ini, except for maxmovielength which is 12000.  It's easy to tell using neogeo games because the part that repeats is the BIOS and it has basically a white screen.  Used i-fun viewer to locate the end frame and once you start reading the log from that frame you'll see what I mean. 

I'll rerun the test's tomorrow on 3countb and send you the files.  I'll also test some other roms and maybe another version of mame.

[Buddabing]

Using Mame .94 and testing with 3countb.

[donnyj]

Using Mame .94 and testing with 3countb.   I am using what you suggest for the mame.ini, except for maxmovielength which is 12000.  It's easy to tell using neogeo games because the part that repeats is the BIOS and it has basically a white screen.  Used i-fun viewer to locate the end frame and once you start reading the log from that frame you'll see what I mean. 

I'll rerun the test's tomorrow on 3countb and send you the files.  I'll also test some other roms and maybe another version of mame.

I found the problem. When a duplicate frame is found, the match count is not incremented. So, if there is one matched frame, then 100 duplicates, then nine more matched frames, the loop will trigger but will be 100 frames too long.

Changing this to increment the match count even when there are duplicate frames could lead to some false positives. To compensate, I suppose I could not increment the match count, but also count how many duplicates there are during the match, then subtract those out of the loop at the end.

I'll have to experiment both ways. Thanks to 2600 for finding this.

Great News!!!!!

On a lot of movies, if the start is 100 frames too late, then the end goes 100 frames too long which creates a loop (from wrong start to wrong end)...  is this problem related?

[2600]

Using Mame .94 and testing with 3countb.   I am using what you suggest for the mame.ini, except for maxmovielength which is 12000.  It's easy to tell using neogeo games because the part that repeats is the BIOS and it has basically a white screen.  Used i-fun viewer to locate the end frame and once you start reading the log from that frame you'll see what I mean. 

I'll rerun the test's tomorrow on 3countb and send you the files.  I'll also test some other roms and maybe another version of mame.

I found the problem. When a duplicate frame is found, the match count is not incremented. So, if there is one matched frame, then 100 duplicates, then nine more matched frames, the loop will trigger but will be 100 frames too long.

Changing this to increment the match count even when there are duplicate frames could lead to some false positives. To compensate, I suppose I could not increment the match count, but also count how many duplicates there are during the match, then subtract those out of the loop at the end.

I'll have to experiment both ways. Thanks to 2600 for finding this.

After thinking about it and rereading your post, I see what you mean by false positives.  Not sure if this is a good solution, but instead of keeping count of the duplicates could you track the first matched frame.  Then you would make the video end with matchframe-1.  Offhand I can't think of any downsides, but maybe you can.

[Kook]

This is a great tool. Nice job Bud. I do have one feature request... when using the automatic loop detection, it would be great if I could specify that I want my saved movies to start in the middle of the loop. Know what I mean? So if I have this option turned on, and let's say the detected loop consists of 300 frames, the AVI will have 300 frames, but the starting frame would be 151 and the ending frame would be 150. Thus when I use the movies as previews in my frontend, and I am scrolling up and down through my list of games, I will see mostly gameplay action in the movie previews instead of intro screens.

Kook

[Silver]

Thats an interesting idea..... Might look a bit strange starting a random point.

I imagine it would be a lot easier to impliment simply starting the movie from a set point - a fraction of the way into the loop or something. ie miss out the first bit altogther... Not quite as nice, but if you are only after a quick glimpse (and some of the movies are very long) it may serve your purpose....

Having it loop round perfectly would probably require a load of file renaming or building of a script/list.....

[Buddabing]

I've uploaded a test version which may address the "some videos are too long" issue.

This zip is 0.94u1, binary only.

I implemented -movieloops 4 for Kook. Basically, it detects the loop as normal, then takes half the length of the loop and keeps capturing frames until that many frames are captured.

So, the movie will contain the second half of the first loop and the first half of the second loop.

Another way to do it would be to shuffle the filenames around. For example, if the loop ran from frame 100 to frame 199, the filenames of the frames would be zoo00100 to zoo00199 (for zookeeper). Then the program would have to rename the files so that zoo00100 would be renamed zoo00150, zoo00101 would become zoo00151, and so forth, and zoo00150 would become zoo00100, zoo00199 would become zoo00149, etc. I chose not to implement it this way for simplicity, but it's the more accurate approach.

<OfficeSpace>Yeeeeah, if you guys would do some tests and see if it runs better, that would be great.</OfficeSpace>

If it does run better for you, I'll go back and change 0.92 and the official source.

[2600]

Speaking of scripts, if anyone is interested I was messing around and have a script that adjusts the start and/or end times.  For example, I run the script in my batch file to remove the neogeo bios screens before virtualdub processes the files.

Let me know if anyone's interested and I'll put it in another thread.

[2600]

Buddabing,
Did a quick test with the test release with the usual 3countb.  The video isn't too long, but something doesn't appear right.

If you take the output,(this changes between runs so don't worry if yours is different):
10-frame match detected between frame 316 and 10194
movie will run between frames 307 and 10185

A: The 10-frame match value is different than the actual run. You'd expect that for the second value under this circumstance, but the begining doesnt' make sense.

B: If you compare the png's the movies is clearly not being made between the two points it says the movie will run between. Looks like the "movie will run between" message is 10 frames too many at the begining and end.  And a quick look at my .vdb file, the addrange is 10175.

[papaschtroumpf]

Speaking of scripts, if anyone is interested I was messing around and have a script that adjusts the start and/or end times.  For example, I run the script in my batch file to remove the neogeo bios screens before virtualdub processes the files.

Let me know if anyone's interested and I'll put it in another thread.

I'm interested. Since I use the videos in mamewah, all the neo-geo games look the same for the first few seconds, whereas the goal is to give the person scrolling down the game list an idea of what gameplay will look like.

[papaschtroumpf]

Buddabing, once you're happy with the fix to the extra frames, can you make a patch for 0.92 (or tell us what to change in what file, I assume it's a pretty localized change).

[Buddabing]

Buddabing,
Did a quick test with the test release with the usual 3countb.

[2600]

Papa, I promise I will work this change into 0.92.

Sorry, burst into laughter when I saw this.


Got two additional question. 
1. What about the 10-frame match message?
2. You have to fill us in, which method did you implement for this?

[Buddabing]

Papa, I promise I will work this change into 0.92.

Sorry, burst into laughter when I saw this.


Got two additional question.

[2600]

The way the match algorithm works now, you can have a 10-frame match between frame 100 and frame 200, if the other 90 frames are duplicates. I did not change the "X-frame match detected" message. For clarity, I can change the message to be "X consecutive non-duplicate frames detected between A and B" if you like.

Don't worry I misunderstood what it was saying.


Checked out the new binary and it worked for me. 
Sorry just got your email, let me know if you still need anything.

[Kook]

Less than 24 hours since I asked, and it now does what I asked for! And it works nicely. I did some testing with -movieloops 4. Sure enough, it figures out where the middle of the movie loop is and starts there (instead of at the beginning of the loop). Way to go Buddabing! This is a nice feature when you want to use the movies as previews in your front end. Thus when you are surfing through a list of games in your front end, you only need to look at each preview for a couple seconds to get an idea what the gameplay is like. This is not to say that all videos come out perfect. For example, some games don't show gameplay during their attract sequences, so obviously Movie Maker can't capture it with automatic loop detection and movieloops -4. But hey, this is the best solution I've seen so far.

It would be even better if Movie Maker accomplished the "start in the middle" thing by renaming PNGs before launching VirtualDUB. The movie making process would be faster (I think it would save an average of about 60-90 seconds in generation time per video with my settings) and in some cases the loops would be more perfect. The numbering thing shouldn't be as complicated as you described. Lets say the loop consists of frames blah00100 to blah00199. First, if there are PNGs you don't need (like blah00000 to blah00099) you would erase them, then simply rename the frames in the FIRST HALF of the loop (100 becomes 200, 101 becomes 201, etc). You end up with blah00150 to blah00249. Then just launch VirtualDub. Of course I'm probably totally out to lunch on that.

At any rate, I'm a happy camper now.

Kook

[Buddabing]

Less than 24 hours since I asked, and it now does what I asked for! And it works nicely. I did some testing with -movieloops 4. Sure enough, it figures out where the middle of the movie loop is and starts there (instead of at the beginning of the loop). Way to go Buddabing! This is a nice feature when you want to use the movies as previews in your front end. Thus when you are surfing through a list of games in your front end, you only need to look at each preview for a couple seconds to get an idea what the gameplay is like. This is not to say that all videos come out perfect. For example, some games don't show gameplay during their attract sequences, so obviously Movie Maker can't capture it with automatic loop detection and movieloops -4. But hey, this is the best solution I've seen so far.

It would be even better if Movie Maker accomplished the "start in the middle" thing by renaming PNGs before launching VirtualDUB. The movie making process would be faster (I think it would save an average of about 60-90 seconds in generation time per video with my settings) and in some cases the loops would be more perfect. The numbering thing shouldn't be as complicated as you described. Lets say the loop consists of frames blah00100 to blah00199. First, if there are PNGs you don't need (like blah00000 to blah00099) you would erase them, then simply rename the frames in the FIRST HALF of the loop (100 becomes 200, 101 becomes 201, etc). You end up with blah00150 to blah00249. Then just launch VirtualDub. Of course I'm probably totally out to lunch on that.

At any rate, I'm a happy camper now.

Kook

I implemented the renaming method instead of the run another loop method. It's faster and more accurate. It's also another 100 lines of code, but who's counting?

What it does is takes the bla00100 and swaps it with bla00150, bla00101 swaps with bla00151, etc, up to bla00149 swapping with bla00199. If the loop has an odd number of frames, lets say from frame 100 to 200 in your example, then frame bla00100 overwrites bla00200 at the end. I don't believe it matters since I believe that VirtualDub uses every other frame anyway.

If you guys want I can implement a -movieskipframes option which would skip a specified number of frames at the beginning of the loop. So if you know the neogeo screen is 150 frames you could specify -movieskipframes 150. The -movieskipframes would be useful if you filter the list beforehand to include only neogeo games. Wouldn't this be easier than a script?

[2600]

Buddabing, if you want to add the movieskipframes that would be great.

It would be a lot easier because the script that I wrote has to use the setrange time which I don't think is as accurate as the frames.  At least not the results I'm getting.  I was going to post the script before and decided against it and was in the middle of posting it again, but I think it is just too hard for a casual user to use easily.

[Kook]

I implemented the renaming method instead of the run another loop method. It's faster and more accurate. It's also another 100 lines of code, but who's counting?

Cool. Is there a binary someplace I can download to try it out?

Kook

[Buddabing]

Buddabing, if you want to add the movieskipframes that would be great.

It would be a lot easier because the script that I wrote has to use the setrange time which I don't think is as accurate as the frames.

[Kook]

I did some quick testing with the test binary and -movieloops 4. It seems that the audio in the avi is out of sync with the video. Specifically, it seems that the video frames are correctly rearranged, but the audio is un-rearranged. I hadn't noticed this or thought about it before. Is it fixable?

 
Kook

[Buddabing]

I did some quick testing with the test binary and -movieloops 4. It seems that the audio in the avi is out of sync with the video. Specifically, it seems that the video frames are correctly rearranged, but the audio is un-rearranged. I hadn't noticed this or thought about it before. Is it fixable?

 
Kook

Oh boy. I didn't think about that. I don't know a way to fix it. I'll have to turn off the audio if -movieloops 4 is selected.

[2600]

The new movieskipframes worked for me.  I used 464 for neogeo games if anyone else tries it.

Thanks Buddabing

[Kook]

Oh boy. I didn't think about that. I don't know a way to fix it. I'll have to turn off the audio if -movieloops 4 is selected.

My vote would be to leave the audio turned on (even if it is out of sync) and maybe make a note about that at the beginning of this thread. It's better than no audio. For me it mostly just means that when you are surfing through a list of games and looking at the video previews, you'll see gameplay video and hear the game's intro music. It's not that big a deal. OR, you could have -movieloops 4 mean "start in the middle with audio out of sync" and -movieloops 5 mean "start in the middle with no audio". Ha! There goes another 100 lines of code :-)

Kook