Request for new feature: Autocrop detection

[evilpenguin]

I'm working on adding features to a video compression plug-in for SageTV and one of the things I really want to do is autocrop letterboxed 4:3 shows and 4:3 shows crammed into 16:9.

I've been using mplayer which has an autocrop detection feature but i've noticed that it doesn't work very reliably with analog recorded SDTV and gets very confused by logos that are partially/entirely outside of the actual video.

Then, I thought, "hey isn't comskip already awesome at that"? So I fired up the GUI and sure enough it's keeping a nice blue box around the actual video at all times. So I was hoping that it would be possible for you to output some type of crop data?

[erik]

There is already a simple solution (at least for me)
Turn on
output_framearray=1

This will make comskip generate a large .csv file containing (the headings minX, maxX are not in the file comskip generates but I inserted them for clarity)

Code: Select all

frame	brightness	scene_change	logo	uniform	sound	minY	MaxY	ar_ratio	goodEdge	isblack	cutscene	minX	maxX		
1	19	0	0	0	276	10	498	141	0	0	0	10	698	0	0
2	30	475	0	11387	272	10	506	141	0	0	100	10	706	85	2112
3	63	265	0	23363	255	10	566	126	0	0	100	10	710	178	2347
4	63	470	0	23281	283	10	566	126	0	0	100	10	710	178	2381
5	63	475	0	23137	263	10	566	126	0	0	100	10	710	179	2346
6	62	475	0	23055	254	10	566	126	0	0	100	10	710	181	2361
7	62	475	0	23048	262	10	566	126	0	0	100	10	710	179	2365
8	62	480	0	23002	257	10	566	126	0	0	100	10	710	178	2379
9	62	475	0	23030	241	10	566	126	0	0	100	10	710	180	2372
10	63	475	0	23074	266	10	566	126	0	0	100	10	710	185	2347
11	63	475	0	23022	264	10	566	126	0	0	100	10	710	177	2434

As you can see border is set to 10 so the outmost 10 pixels are discarded
You can change this by changing the border setting in the ini file

The minY, maxY, minX and maxX value can be used to determine the crop window

[evilpenguin]

I might be able to make that work for me. I can run comskip for ~1-2 minutes, send it a ctrl+c, and then mine the csv file for data.

Is there anyway I can configure it to only output the csv file? I wouldn't want to blow away peoples existing comskip edl or txt files.

[erik]

set
output_default=0
and
output_edl=0
but do check if this does not result in deleting a previously present .txt or .edl file as I never have tested this

[evilpenguin]

The data in the csv file is perfect, i've got a nice, efficient, function reading in data and outputting the 100% correct crop data everytime.

A few things,

1) I don't think the output folder parameter in the CLI is working properly. Right now, no matter what I put its outputting the csv file next to the original video. That is unless I'm completely missing something... which has been known to happen.

Code: Select all

"D:/perl2dvd Dev/Development/perl2dvd/comskip.exe" --ini="D:/perl2dvd Dev/Development/perl2dvd/comskip.ini" "E:/SageRecordings/GoodEats-StuffIt-849894-0.mpg" "D:/perl2dvd Dev/Development/"

2) Is there anyway to get it to output the .csv file while its running rather than the end. I'm having trouble sending it a clean "ctrl+c" but I can just outright kill the process.

[erik]

The output folder is only used for the generated cutlists.
However your remark does make sense

W.r.t. the crl+C or outputting the .CSV while running.
As audio is often out of sync with video I can not output the .csv while the frames are being processed. Will have too look for the crtl+C thing. How to catch it etc...
Don;t expect quick progress.

[evilpenguin]

I've been doing some brainstorming and I'm fairly certain that I can work around both of those issues pretty easily.

Right now i'm thinking that I can just generate 5 evenly spaced 20 sec clips from the first 30 minutes of the video, put them in my temp folder, run comskip on them, concat the data, and pull out the data like that.

[erik]

At the end of the log file comskip reports the dominant aspect ratio

Code: Select all

Logo fraction:              0.5659      
Maximum volume found:         3030
Average volume:                970
Sound threshold:               500
Silence threshold:             100
Minimum volume found:            0
Average frames with silence:    78
Black threshold:                20
Minimum Black level found:      15
Average brightness:           1064
Uniformity level:              260
Average non uniformity:        446
Maximum gap between logo's:    268
Suggested logo_threshold:   0.7750
Suggested shrink_logo:	         5
Max commercial size found:     109
Dominant aspect ratio:      1.2600
Score threshold:            1.0500
Framerate:                  25.00
Total commercial length:    0:20:43.00

[evilpenguin]

I've been doing some brainstorming and I'm fairly certain that I can work around both of those issues pretty easily.

Right now i'm thinking that I can just generate 5 evenly spaced 20 sec clips from the first 30 minutes of the video, put them in my temp folder, run comskip on them, concat the data, and pull out the data like that.

I've got this method coded up and its working amazingly well. Very fast and accurate across a wide range of sources. Also, as an added bonus it works for non-mpegs as it can transcode them if necessary.

So I guess in addition to being the best commercial detecting software, comskip is the most reliable crop detection software as well. I wonder what else it can do

[erik]

Thanks for the kind words!

And of course comskip can do even more, such as demux.....

[gEd]

Generated using Comskip 0.79.113

My Dominant aspect ratio is always zero. Do I need to change a setting to get it to show the correct value. I'm trying to use statik's latest encoding bat file.

cheers

[edit] I just tried it with the UK ini file that was posted in the tuning forum and I now have a DAR value. cool!

[erik]

The next build can output a .aspects file with this content

Code: Select all

0:00:00.01 1280x 720 1.87 minX=  10, minY=  26, maxX= 910, maxY= 502
0:00:00.05 1280x 720 1.80 minX=  10, minY=  10, maxX=1270, maxY= 710
0:04:28.91 1280x 720 1.30 minX= 186, minY=  10, maxX=1094, maxY= 710
0:05:00.35 1280x 720 2.30 minX=  10, minY=  82, maxX=1270, maxY= 630
0:05:20.37 1280x 720 1.80 minX=  10, minY=  10, maxX=1270, maxY= 710
0:06:13.94 1280x 720 1.38 minX= 162, minY=  10, maxX=1118, maxY= 710
0:06:29.07 1280x 720 1.80 minX=  10, minY=  10, maxX=1270, maxY= 710
0:13:36.41 1280x 720 1.38 minX= 162, minY=  10, maxX=1118, maxY= 710
0:15:00.73 1280x 720 1.80 minX= 162, minY=  82, maxX=1110, maxY= 630
0:15:20.22 1280x 720 1.38 minX= 162, minY=  10, maxX=1118, maxY= 710
0:16:36.73 1280x 720 1.80 minX=  10, minY=  10, maxX=1270, maxY= 710
0:24:04.34 1280x 720 1.38 minX= 162, minY=  10, maxX=1118, maxY= 710
0:25:14.14 1280x 720 2.05 minX= 170, minY= 122, maxX=1118, maxY= 590
0:25:17.75 1280x 720 1.80 minX= 162, minY=  82, maxX=1118, maxY= 630
0:25:19.53 1280x 720 1.38 minX= 162, minY=  10, maxX=1118, maxY= 710
0:25:41.40 1280x 720 1.64 minX= 170, minY=  66, maxX=1118, maxY= 646
0:25:43.51 1280x 720 1.73 minX= 170, minY=  82, maxX=1118, maxY= 630
0:25:47.63 1280x 720 0.00 minX= 154, minY=  10, maxX=1118, maxY= 710
0:25:50.88 1280x 720 1.38 minX= 154, minY=  10, maxX=1118, maxY= 710
0:27:40.71 1280x 720 1.80 minX=  10, minY=  10, maxX=1270, maxY= 710
0:33:44.52 1280x 720 1.38 minX= 162, minY=  10, maxX=1118, maxY= 710
0:37:14.78 1280x 720 1.02 minX= 282, minY=  10, maxX= 990, maxY= 710
0:37:15.81 1280x 720 1.38 minX= 154, minY=  10, maxX=1118, maxY= 710
0:37:34.17 1280x 720 1.80 minX=  10, minY=  10, maxX=1270, maxY= 710
0:45:27.37 1280x 720 1.38 minX= 154, minY=  10, maxX=1118, maxY= 710
0:46:44.80 1280x 720 1.80 minX=  10, minY=  10, maxX=1270, maxY= 710
0:47:45.06 1280x 720 1.38 minX= 162, minY=  10, maxX=1118, maxY= 710
0:49:21.76 1280x 720 1.80 minX=  10, minY=  10, maxX=1270, maxY= 710
0:59:39.52 1280x 720 1.38 minX= 154, minY=  10, maxX=1118, maxY= 710

Is this usefull?