MAME Movie Maker released

[Buddabing]

Incidentally - hope I'm not confusing things - but I've just run colmns97 through and it found a loop and made a nice video no problem. This is with standard settings, using Buddabings own mame 0.89 compile dated 11th Jan 2005. Not sure if that means the error is in 0.90 only or is something else......

EDIT: A rerun with identical settings with the latest 0.90 (your build) produces the " image 1248 resolution (704x480) does not match the 1st image (302x240)" error in vdub. Looks like the resolution work around is not overwriting the earlier png's in the newer build?

When I added the moviesnap_directory parameter, it broke the resolution change detection code. I'll fix it today.

[rerstad]

Great! I ran it all last night and the only error I got was in vdub when the resolution changed (I also changed batch.bat to clearly delete the files in moviesnp after making each movie--that file was getting so big cumulatively that it would take forever for me to even open it, deleting the files seems to have helped some (of course, it could just be me)).

[Howard_Casto]

since you asked for examples, some games that don't detect for me

64th street

arabian fight

(virtually the entire "fighter" genre as filtered by the catver)

I'm still thinking that the transition method of many games is what is killing the loop detection code.

Let me give you some examples:  (no I haven't tested these games with the movie maker, they are just examples)

Robocop....  it has a startup cinematic sequence in which the screen is wiped a very very dark grey (It looks black if you aren't paying attention) 4 distinct times.  Then it has a random bit of gameplay in the attract mode, and again, at the start and the end of this sequence, there's another grey wipe, creating at least one solid grey frame. 

Nearly every kof game (bah!!!)  has a seizure-enducing intro that contains several solid color wipes (mostly black and white).  They occur enough times that the code would get confused. 

If it were a few isolated games, then I wouldn't mention it, but sadly nearly every game produced from 1991 on has these type of screen transitions. 


The only way I can think to fix this issue is to somehow check the frames and make sure they aren't a solid color field, rejecting any loops between solid color frames.  Where pngs aren't palletted though, I have no clue how you would do it quickly and accurately enough.   The only thing I can think of is the file size.  The fewer the colors an image has, the smaller it is. 

Sorry I can't think of a solution.  I guess knowing why it isn't working is better than nothing though. 

[rerstad]

Oddly enough, despite any other problems I've had running this, mine produced a movie for Arabian Fight last night.  I haven't looked at it to see what's in it, I just remember seeing it run and know that it completed, so it must have detected. . . .

[sWampy]

Didn't someone say there was a database of start/end frames for games?

[Buddabing]

I have uploaded a new executable which should correct the resolution change. I'm creating a video now for colmns97. It looks fine so far. It's taking forever. More than an hour to generate the pngs and another hour for the encoding. But, it detected the demo loop perfectly.

Howard: Thank you for your examples. I've never played any of those games so I don't know anything about their graphical
transitions, but I'll take a look at them.

Swampy: There is such a database. I probably could integrate that start-stop thing with this code. Maybe a movieloops 2 option would use that file.....

Rerstad: I believe Howard was referring to the loop detection, not it making a movie. I haven't looked at those movies either, but they probably run into the max frame limit.

EDIT: The problem with the file (provided by 2600 on the first page of this thread) is that it has the starting frame of the movie but not the end frame!

[donnyj]

question about cbnj, it creates a 44 byte .wav file that can't be played so virtualdub errors out, please confirm???

edit: seems to create correct wav file after the countdown finishes, but what about before (i.e. 10,000 or less frames?)

[vibez]

how do you make listgen create the makemovies.bat. All I'm getting is  a listxml.txt & output.txt

[rerstad]

The latest build of v.90 seems to have taken care of all of the problems.  I've been running it for about 14 hours straight with no errors and no stops.  I haven't looked at each movie, but it keeps plugging away . . . .

[vibez]

I'm getting a 'error creating directsound: 887800a' message. any ideas what this for?

[Silver]

I'm getting a 'error creating directsound: 887800a' message. any ideas what this for?

Is this an error from mame or virtualdub?

can you run the game normally with mame? (just type 'mame romname' at the command line)

[Buddabing]

how do you make listgen create the makemovies.bat. All I'm getting is  a listxml.txt & output.txt

Edit your listgen.ini file so that you have:
frontend moviebatch
output makemovies.bat

Or:
listgen -frontend moviebatch -output makemovies.bat

You must have a filter.sql and the mamepath field in listgen.ini must be valid.

For your sound issue, try -movieaudio 0 and see if that makes any difference.

That 'error creating directsound' error is a MAME error.

HTH,
Buddabing

[Lilwolf]

Ok, this is probably a pretty stupid question / suggestion.

but wouldn't it be easier to scrap the loop detection and add a key for start / stop... Then record all new changes to a file......

Then have a setting to import new save files...

IE, everyone who uses it does the start/stop to 100 games (an hour)... get 20 people doing it and you could have all games start / stop time / frame... And then just do it...

I think the loop detection is pretty !@# sweet...  but the number of exceptions also seem pretty high...

[Buddabing]

Ok, this is probably a pretty stupid question / suggestion.

but wouldn't it be easier to scrap the loop detection and add a key for start / stop... Then record all new changes to a file......

Then have a setting to import new save files...

IE, everyone who uses it does the start/stop to 100 games (an hour)... get 20 people doing it and you could have all games start / stop time / frame... And then just do it...

I think the loop detection is pretty !@# sweet...

[donnyj]

Great idea, I am willing to participate HEAVILY.  I have another suggestion.  Pick a desired frame # per game and it will save a "title" snap.  The existing ones (for snap, title, etc.) seem terrible quality and several haven't been updated, so they just show the old non working screen (i.e. colorbars).

----

separately:

Aquarush seems really odd to me.....,  the command line SAYS 640x480, but it displays and take snaps at 640x240.  (Smashed) ?  If it is a mame problem, then a resize command needs to be added to virtual dub???


AWESOME PROJECT, I am working hard on my end breaking it and seeing how it works...  Would GLADLY contribute within my realm of abilities (i.e., no programming)

[donnyj]

ok, this will be more difficult than I first thought 

Will have to pick the frame numbers from the PNGs, NOT from the AVI's correct?

Also, is ???00000.png  frame 1 or is ???00001.png?

What is the prefererred video?  Just the attract sequence (before or after stuff like "NEO GEO" logo)?  Or is it from power on until it starts to repeat back to some point again?  In other words, until it starts to actually repeat?

Or do you wait for a repeat and start from there until it repeats again?

[Buddabing]

ok, this will be more difficult than I first thought

[donnyj]

Ok, I am starting from ZZYZZYX and moving towards A

Noting, start, stop, and title frames


My hope is that when i have this list, I can feed it back into MMM and it will return the AVI's.... ?

Because right now, I am just creating pictures only, skipping both audio and avi creation.....

Definately a slow and tedious process 

[Silver]

Hey good plan - keep posting with your progress to make sure people don't repeat work....

Also, it may be worth establishing some guidelines about where to start stop attracts - eg some games have a several minute loop - should you take it all or just a shorter attract? etc..etc...

I suggest you save your work in a simple csv txt file with something like:

romname,startframe,endframe
romname,startframe,endframe.....

to make it nice and easy to use/merge later on.

[Buddabing]

Hey good plan - keep posting with your progress to make sure people don't repeat work....

Also, it may be worth establishing some guidelines about where to start stop attracts - eg some games have a several minute loop - should you take it all or just a shorter attract? etc..etc...

I suggest you save your work in a simple csv txt file with something like:

romname,startframe,endframe
romname,startframe,endframe.....

to make it nice and easy to use/merge later on.

Agreed. Unless anyone has any objections, I'll modify ListGen to read this CSV and spit out a new batch file which will call batch.bat with the start and end frames, if they are in the CSV, or with the original parameters, if not.

This requires adding some % parameters to batch.bat: a movieloops parameter (1 or 2 depending on if the game is in the CSV), a moviestartframe parameter, and a movieendframe parameter.

Donny - I would suggesting starting with the games in the ssf.txt file and working backwards like you are doing.

Regards,
Buddabing

[Silver]

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.

[Buddabing]

For anyone looking to find start and end frames, I have a tool I find extremely useful.

It's called I-fun viewer from Xequte Software:

Here's a link to download

It shows a scrolling list of thumbnails of the png snapshots, with the currently selected snapshot shown larger. It can scroll through hundreds of images very quickly.

I highly recommend it, and the price is right.

[2600]

Hey good plan - keep posting with your progress to make sure people don't repeat work....

Also, it may be worth establishing some guidelines about where to start stop attracts - eg some games have a several minute loop - should you take it all or just a shorter attract? etc..etc...

I suggest you save your work in a simple csv txt file with something like:

romname,startframe,endframe
romname,startframe,endframe.....

to make it nice and easy to use/merge later on.

I see you all are still hard at work. Great work all of you. Since these are attract mode movies, I vote for the whole thing.   Seems much more official that way.  This would mean several loops for some games.

[Silver]

I vote for the whole thing.   Seems much more official that way.  This would mean several loops for some games.

I'm in.....

People can always change the file to individual taste, but we can have an offical file for people to start with....

[donnyj]

suggestions for listgen:

put the default makemovies.bat "script" in the filters.sql file as an example

do the syntax checking BEFORE loading all the data files (so errors bomb out faster)

question:

how would I reverse the sort order?  for example, sort by year in reverse?

[papaschtroumpf]

I didn't think the problem was detecting unique frames, MMM probably does something similar (haven't looked at the code). The problem is that there can be 2 identical frames even though the loop isn't finished, such as a fully black frame during a "fade to black".

[Howard_Casto]

I didn't think the problem was detecting unique frames, MMM probably does something similar (haven't looked at the code). The problem is that there can be 2 identical frames even though the loop isn't finished, such as a fully black frame during a "fade to black".

close.... the detector has some common sense built in if I'm reading the code properly, so a single fade to black won't screw it up.  Unfortunately two or more fades to black (or any color for that matter)  will, and since many games do this we are kinda screwed at this point. 

[Buddabing]

suggestions for listgen:

put the default makemovies.bat "script" in the filters.sql file as an example

do the syntax checking BEFORE loading all the data files (so errors bomb out faster)

question:

how would I reverse the sort order?  for example, sort by year in reverse?

Are you talking about syntax checking in the SQL? I might implement something like a "-sqltest" option. Do you really think that's worth the effort it will take?

Reversing the sort order is done with "order by":

select * from gamedata
   order by year desc;

For ascending order use "asc"

select * from gamedata
   order by year asc;

Some of the years are missing so you will have modern games at the end of the descending order list.

[Buddabing]

I didn't think the problem was detecting unique frames, MMM probably does something similar (haven't looked at the code). The problem is that there can be 2 identical frames even though the loop isn't finished, such as a fully black frame during a "fade to black".

Fades to black aren't a problem. If you change the -moviethreshold value, fades to dark gray won't be a problem either.

Fades to a non-black color will cause the loop detection to trigger on the second fade.

I could try figuring the max and min pixel value of each image, and not triggering loop detection if the difference between max and min is less than -moviethreshold. This will prevent a false loop detect of all solid colors where the color is the same, at the cost of considerably more calculation.

When I looked at at Robocop, the movie looped at the high score list. For kof94, you can see the "credit 00" at the beginning and end of the movie on an otherwise black screen. On 64th street, the loop detection happens when the dialogue between the two characters is blanked out, the screen is not blank then.

Kof94 and 64street may be improved by increasing the -minmovielength parameter. Otherwise, someone can go and manually get the start and end frame and generate a movie from that.

The coding is complete on the next version which will have the -moviestartframe and -movieendframe parameter. Using those parameters may make the movie making process a bit faster.

Now I have to go and port everything over again now that .91 is out.

[papaschtroumpf]

Now I have to go and port everything over again now that .91 is out.

I could spend my entire spare time upgrading rather than playing if tried to keep up with MAME. In fact the only reason I upgraded from 0.88 is MMM, now I may have to do it again? Aaarg!
I haven't looked at the code and patches so I'm not sure if I can keep up with MMM by diffing the code or I'm better off upgrading to 0.91 when you release MMM for 0.91.

I guess I could probably jsust upgrade MAME itself without worrying about keeping the ROMS, screenshots, flyers, etc... in sync with the MAME version, since that's where I spend 99% of the upgrade time.

[Buddabing]

Now I have to go and port everything over again now that .91 is out.

I could spend my entire spare time upgrading rather than playing if tried to keep up with MAME. In fact the only reason I upgraded from 0.88 is MMM, now I may have to do it again? Aaarg!
I haven't looked at the code and patches so I'm not sure if I can keep up with MMM by diffing the code or I'm better off upgrading to 0.91 when you release MMM for 0.91.

I guess I could probably jsust upgrade MAME itself without worrying about keeping the ROMS, screenshots, flyers, etc... in sync with the MAME version, since that's where I spend 99% of the upgrade time.

Yeah, I hear you, which is one reason my cab still runs 0.78.

It turns out that the key file that most of the MMM code is in (common.c) didn't change from version 0.90 to 0.91. So it will be very simple to port MMM. The interface code to CPMaker is in artwork.c, which didn't change either.

Long story short, look for "BuddaMAME" v0.91 very soon.

[donnyj]

suggestions for listgen:

put the default makemovies.bat "script" in the filters.sql file as an example

do the syntax checking BEFORE loading all the data files (so errors bomb out faster)

question:

how would I reverse the sort order?

[Buddabing]

I have uploaded a new version of MMM which supports the -moviestartframe and -movieendframe parameters.

When using -moviestartframe and -movieendframe, MAME ran at 38.5 fps when capturing the frames for defender. Using the original loop detection, MAME ran at 34.7 fps. That's about 10% faster, your mileage may vary.

They've changed the compile environment so that version 0.91 of MAME won't compile anymore on my machine. So MMM 0.91 will be delayed somewhat.

[Silver]

I've been playing around a fair bit with png's and mng's recently. I was wondering if it is easy to alter the zlib compression level used when mame creates a png image?

[Buddabing]

I've been playing around a fair bit with png's and mng's recently. I was wondering if it is easy to alter the zlib compression level used when mame creates a png image?

Yes, in fact I did so when I was investigating calculating the MD5 from the image data instead of the image file.

Here is the relevant code from png.c:

int png_deflate_image(struct png_info *p)
{
   unsigned long zbuff_size;

   zbuff_size = (p->height*(p->rowbytes+1))*1.1+12;

   if((p->zimage = (UINT8 *)malloc (zbuff_size))==NULL)
   {
      logerror("Out of memory\n");
      return 0;
   }

[Silver]

aha thanks... I presume the default is around 5?

If its no hassle to you, it would be very handy (to me) if you could add an option as described - in particular both disabling the compression completely (uncompressed png's - large I know) and max compression.

My compiling environment for mame is still out of data from the last time they updated.... need to sort that!

[Buddabing]

aha thanks... I presume the default is around 5?

If its no hassle to you, it would be very handy (to me) if you could add an option as described - in particular both disabling the compression completely (uncompressed png's - large I know) and max compression.

My compiling environment for mame is still out of data from the last time they updated.... need to sort that!

Okay, done.

-moviecompression now reflects compression used in the zlib. Allowable values are -1 (default, which is currently 6 but is the system's compromise between speed and compression) or 0-9.
0 = no compression
1 = fastest
9 = best compression

I ran a quick test on defender with -moviestartframe 222 and -movieendframe 888. A proper test would make many more png images of several different games.

X       fps          png size
-1     42.93      6799933
0      45.08      18014856
1      43.51      6851916
2      42.33      6850810
3      43.88      6850060
4      43.21      6814655
5      42.67      6809736
6      43.45      6799933
7      40.98      6799592
8      43.02      6793833
9      23.46      6787289

Don't know why -1 and 6 took different amounts of time.

[Silver]

Thanks for that, much appreciated!

[2600]

Just a note, but the original ssf.txt was used for an entirely different purpose so use it as a guide not an absolute truth.

That being said I'm not having much of a problem with loop detection for the most part.  Any that happen to not get detected or detected incorrectly, I change the minmovielength and sometimes the maxmovielength and the loop is detected no problem.  Not a big fighter fan, but just did the KOF series and MK series and verified the loop was detected properly with no problem.  Will try the others mentioned above.  Recently, I've been running the batch and then outputting the messages to a txt file.  With some creative search and replace or a script you get the romname,startframe,endframe.

Silver, do you still want to maintain the list?  Is PMing you the results ok or would you rather have it sent to another address.

[Silver]

Just a note, but the original ssf.txt was used for an entirely different purpose so use it as a guide not an absolute truth.

Yes I knew that - I was using the ssf for its oringinal purpose, but thought it would make a good starting point!

Recently, I've been running the batch and then outputting the messages to a txt file.  With some creative search and replace or a script you get the romname,startframe,endframe.

Execellent - I was actually going to post and suggest this, especially if you are going to the trouble of adjusting the settings for some games. Most of the work for this can be done by 'BuddaMame' and just thoses that are problematic manually. If you've saved all the vdb files or messages, thats fantastic - you can just zip them up and email if thats easier, or Pm'ing is fine.
I've actually thrown together a Mysql database on the web with all the 0.91 roms - partually just for the learning experience for sql/php - although I currently have a nasty "No OS" situation so thats not up and runnning yet.

Silver, do you still want to maintain the list?  Is PMing you the results ok or would you rather have it sent to another address.

Basically yes, in anyway you like, to Silver@silverfoxy.plus.com

Thanks