Frame/gop accurate video splitting (fps, gop, length known)
Hey there! A simple use case: 1) a sample mp4 file of 23.98 tbr, 264 + aac, 1:51.27 in length 2) the file (lotr trailer) encoded with: ./ffmpeg -i ./lordoftherings_h640_ml.mov -vcodec libx264 -g 50 ./lord.mp4 3) we want to split it in 3 parts: 30s, 10s and the rest of the stream Chunks we want to have: 1) ~30s (00:00:00 - 00:00:30) 2) ~10s (00:00:30 - 00:00:40) 3) ~71s (00:00:40 - 00:01:51) File info: Duration: 00:01:51.27, start: 0.000000, bitrate: 256 kb/s Stream #0.0(eng): Video: h264, yuv420p, 640x272, 192 kb/s, 47.90 fps, 23.98 tbr, 24k tbn, 47.95 tbc Stream #0.1(eng): Audio: aac, 48000 Hz, stereo, s16, 58 kb/s The only tools right now are -vframes, -ss and -t. Frame-accurate seeking is missing/broken (?). Below is my approach. a) chunk 1 is 30s, so with 23.98 fps that gives 719.5 frames. b) Because it's gop 50 we round down to 700 frames = 29.1909925 seconds from start c) Chunk 1 extraction: "-vframes 699" d) Chunk 2 starts from frames 700 and ends in 10s = 239.8 frames ~ 250 frames (gop 50) e) Chunk 2 extraction: "-ss 29.1909925 -vframes 250" f) Chunk 3 starts at 40s = 959.2th frames (23.98 fps). Rounding down to 50 gives us 950th frame g) 950th frame is at aproximately 39.61635 sec. g) Chunk 3 extraction: "-ss 39.61635" Of course I wouldn't be bothering the community if it worked properly :-) The chunks are off by many seconds, lengths are off. Please assist - what am I doing wrong? Is there any more elegant way to do that? Also: a simpler case that I thought would "just work" - split the stream in two: ./ffmpeg -y -ss -t 30 -i ./lord.mp4 -vcodec copy -acodec copy lord-chunk1.mp4 ./ffmpeg -y -ss 00:00:30 -i ./lord.mp4 -vcodec copy -acodec copy lord-chunk2.mp4 It doesn't work! This will also end badly with two arbitrary length chunks with seek times that make little sense. How is it done properly? Thanks! Arthur. ps: FFmpeg version SVN-r20880 -- __ /.)\ +48 695 600 936 \(./ abodera at gmail.com
On Mon, Dec 28, 2009 at 12:48 PM, Artur Bodera <abodera at gmail.com> wrote:
Hey there! It doesn't work! This will also end badly with two arbitrary length chunks with seek times that make little sense. How is it done properly?
Bump :-) None splits videos ? Any suggestions are welcomed! A. -- __ /.)\ +48 695 600 936 \(./ abodera at gmail.com
On Wed, Dec 30, 2009 at 1:29 PM, Artur Bodera <abodera at gmail.com> wrote:
On Mon, Dec 28, 2009 at 12:48 PM, Artur Bodera <abodera at gmail.com> wrote:
Hey there!
It doesn't work! This will also end badly with two arbitrary length chunks with seek times that make little sense. How is it done properly?
How to just split a video into 3 parts? -- __ /.)\ +48 695 600 936 \(./ abodera at gmail.com
On Mon, Dec 28, 2009 at 12:48 PM, Artur Bodera <abodera at gmail.com> wrote:
[...] Chunks we want to have: ? 1)? ~30s (00:00:00 - 00:00:30) ? 2)? ~10s (00:00:30 - 00:00:40) ? 3)? ~71s (00:00:40 - 00:01:51)
File info: ? Duration: 00:01:51.27, start: 0.000000, bitrate: 256 kb/s ??? Stream #0.0(eng): Video: h264, yuv420p, 640x272, 192 kb/s, 47.90 fps, 23.98 tbr, 24k tbn, 47.95 tbc ??? Stream #0.1(eng): Audio: aac, 48000 Hz, stereo, s16, 58 kb/s
A small test - the easy way just using -ss and -t: Encoding of chunk 1: ffmpeg -y -t 30 -i lord.mp4 -vcodec copy -acodec copy lord-c1.mp4 result: Duration: 00:00:30.05, start: 0.000000, bitrate: 286 kb/s Stream #0.0(eng): Video: h264, yuv420p, 640x272, 237 kb/s, 47.95 fps, 23.98 tbr, 48k tbn, 47.95 tbc Stream #0.1(eng): Audio: aac, 48000 Hz, stereo, s16, 43 kb/s Encoding of chunk 2: ffmpeg -y -ss 30 -t 10 -i lord.mp4 -vcodec copy -acodec copy lord-c2.mp4 result: Duration: 00:00:15.27, start: 0.000000, bitrate: 226 kb/s Stream #0.0(eng): Video: h264, yuv420p, 640x272, 156 kb/s, 47.95 fps, 23.98 tbr, 48k tbn, 47.95 tbc Stream #0.1(eng): Audio: aac, 48000 Hz, stereo, s16, 64 kb/s Encoding of chunk 3: ffmpeg -y -ss 40 -i lord.mp4 -vcodec copy -acodec copy lord-c3.mp4 result: Duration: 00:01:12.12, start: 0.000000, bitrate: 268 kb/s Stream #0.0(eng): Video: h264, yuv420p, 640x272, 199 kb/s, 47.95 fps, 23.98 tbr, 48k tbn, 47.95 tbc Stream #0.1(eng): Audio: aac, 48000 Hz, stereo, s16, 63 kb/s After re-joining the video has 117.44 s (the original file had 111,27s). The fragments are inaccurate and overlap. -- __ /.)\ +48 695 600 936 \(./ abodera at gmail.com
Duration: 00:00:30.05
Hasn't anyone working on ffmpeg ever heard of timecode? You know, hours, minutes, seconds and frames, as is the standard for timing video in more or less all cases except ffmpeg? Ah, yes, they have: https://lists.mplayerhq.hu/pipermail/ffmpeg-devel/2007-April/027586.html ...but they thought they knew better. Facepalm. P
On Jan 06 11:35:05, Phil Rhodes wrote:
Duration: 00:00:30.05
Hasn't anyone working on ffmpeg ever heard of timecode? You know, hours, minutes, seconds and frames, as is the standard for timing video in more or less all cases except ffmpeg?
Don't know about timecode, but actually with 47.90 fps, a "chunk of length 00:00:10.0" doesn't even necessary correspond to one definite border frame. Right?
On Jan 06 12:47:19, Jan Stary wrote:
On Jan 06 11:35:05, Phil Rhodes wrote:
Duration: 00:00:30.05
Hasn't anyone working on ffmpeg ever heard of timecode? You know, hours, minutes, seconds and frames, as is the standard for timing video in more or less all cases except ffmpeg?
Don't know about timecode, but actually with 47.90 fps, a "chunk of length 00:00:10.0" doesn't even necessary correspond to one definite border frame. Right?
No really, with 47.90 fps and -g 50, how many vframes is the "first 10 seconds" supposed to have?
On Wed, Jan 6, 2010 at 1:34 PM, Jan Stary <hans at stare.cz> wrote:
No really, with 47.90 fps and -g 50, how many vframes is the "first 10 seconds" supposed to have?
Let's simplify the whole thing. Folks, please suggest how can I split the file in two? My attempt: Original file: Duration: 00:01:51.27, start: 0.000000, bitrate: 268 kb/s Command: ffmpeg -t 55 -i lord.mp4 -vcodec copy -acodec copy lord-half1.mp4 Command: ffmpeg -ss 55 -i lord.mp4 -vcodec copy -acodec copy lord-half2.mp4 Result: half1: Duration: 00:00:55.03, start: 0.000000, bitrate: 320 kb/s half2: Duration: 00:00:56.32, start: 0.000000, bitrate: 219 kb/s Length mismatch. In the original file, there is a white blast between the scene with the Black Riders and Gandalf's face. It's missing in the end of half1 and beginning of half2. I've rejoined the files to verify: # ffmpeg -i lord-half1.mp4 -vcodec mpeg2video -acodec copy lord-half1.mpg # ffmpeg -i lord-half2.mp4 -vcodec mpeg2video -acodec copy lord-half2.mpg # cat lord-half1.mpg lord-half2.mpg > lord-joined.mpg The missing scene is more visible now. Normalization attempt: 1. 55'th second is 2637,25 frames (47.95 fps). 2. Rounding down to 2600 frames (gop 50) 3. 2600 frames is close to 54.223149113660062565172054223149 seconds 4. Rounding up to 54.2232 and crossing fingers 5. ffmpeg -t 54.2232 -i lord.mp4 gives us: Duration: 00:00:54.24 6. ffmpeg -ss 54.2232 -i lord.mp4 gives us file: Duration: 00:00:59.20 7. total length: 113.44 original length 111,27 a workaround attempt: 6. The -t 54.2232 gave us duration 54.24 so we will seek directly to it 7. ffmpeg -ss 54.24 -i lord.mp4 gives us file: Duration: 00:00:59.20 7. total length: 113.44 Looks clear! Unfortunatelly the scene where Saruman throws Gandalf in the air is visible twice. -- __ /.)\ +48 695 600 936 \(./ abodera at gmail.com
Artur Bodera wrote:
On Wed, Jan 6, 2010 at 1:34 PM, Jan Stary <hans at stare.cz> wrote:
No really, with 47.90 fps and -g 50, how many vframes is the "first 10 seconds" supposed to have?
Let's simplify the whole thing. Folks, please suggest how can I split the file in two?
My attempt: Original file: Duration: 00:01:51.27, start: 0.000000, bitrate: 268 kb/s Command: ffmpeg -t 55 -i lord.mp4 -vcodec copy -acodec copy lord-half1.mp4 Command: ffmpeg -ss 55 -i lord.mp4 -vcodec copy -acodec copy lord-half2.mp4 Result: half1: Duration: 00:00:55.03, start: 0.000000, bitrate: 320 kb/s half2: Duration: 00:00:56.32, start: 0.000000, bitrate: 219 kb/s
Length mismatch. In the original file, there is a white blast between the scene with the Black Riders and Gandalf's face. It's missing in the end of half1 and beginning of half2.
I've rejoined the files to verify: # ffmpeg -i lord-half1.mp4 -vcodec mpeg2video -acodec copy lord-half1.mpg # ffmpeg -i lord-half2.mp4 -vcodec mpeg2video -acodec copy lord-half2.mpg # cat lord-half1.mpg lord-half2.mpg > lord-joined.mpg The missing scene is more visible now.
Normalization attempt: 1. 55'th second is 2637,25 frames (47.95 fps). 2. Rounding down to 2600 frames (gop 50) 3. 2600 frames is close to 54.223149113660062565172054223149 seconds 4. Rounding up to 54.2232 and crossing fingers 5. ffmpeg -t 54.2232 -i lord.mp4 gives us: Duration: 00:00:54.24 6. ffmpeg -ss 54.2232 -i lord.mp4 gives us file: Duration: 00:00:59.20 7. total length: 113.44 original length 111,27
a workaround attempt: 6. The -t 54.2232 gave us duration 54.24 so we will seek directly to it 7. ffmpeg -ss 54.24 -i lord.mp4 gives us file: Duration: 00:00:59.20 7. total length: 113.44 Looks clear!
Unfortunatelly the scene where Saruman throws Gandalf in the air is visible twice.
IMHO you need to work on video and audio separatedly. Video and audio almost always have a delay to each other. Say audio lags .08 which in your case is about 4 frames. If you say I want 54secs of a movie you'll get 54.08 If you specify -v 2600 you'll get that amount of frames and audio will be a little short. I use this method to transcode first all video from 1source with 3DualCorePC's and then transcode audio part separated and then mux them together to 1encoded file. To make it work you have to know the GOP-size and split at that point and that point only. The -v amount has to be 1less in my experience with PAL DVB sources. It may/could be different on other source material. Hope it helps.
On Jan 06 14:38:14, Artur Bodera wrote:
On Wed, Jan 6, 2010 at 1:34 PM, Jan Stary <hans at stare.cz> wrote:
No really, with 47.90 fps and -g 50, how many vframes is the "first 10 seconds" supposed to have?
Let's simplify the whole thing. Folks, please suggest how can I split the file in two?
My attempt: Original file: Duration: 00:01:51.27, start: 0.000000, bitrate: 268 kb/s
Also, $ ffmpeg -i lord.mp4 FFmpeg version SVN-r21032, Copyright (c) 2000-2010 Fabrice Bellard, et al. built on Jan 6 2010 13:45:52 with gcc 3.3.5 (propolice) configuration: libavutil 50. 7. 0 / 50. 7. 0 libavcodec 52.45. 0 / 52.45. 0 libavformat 52.46. 0 / 52.46. 0 libavdevice 52. 2. 0 / 52. 2. 0 libswscale 0. 7. 2 / 0. 7. 2 Compiler did not align stack variables. Libavcodec has been miscompiled and may be very slow or crash. This is not a bug in libavcodec, but in the compiler. You may try recompiling using gcc >= 4.2. Do not report crashes to FFmpeg developers. Seems stream 0 codec frame rate differs from container frame rate: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 47.95 (48000/ 1001) -> 23.98 (24000/1001) Input #0, mov,mp4,m4a,3gp,3g2,mj2, from 'lord.mp4': Metadata: major_brand : isom minor_version : 512 compatible_brands: isomiso2avc1mp41 encoder : Lavf52.43.0 Duration: 00:01:51.27, start: 0.000000, bitrate: 268 kb/s Stream #0.0(eng): Video: h264, yuv420p, 640x272, 204 kb/s, 47.90 fps, 23.98 tbr, 24k tbn, 47.95 tbc Stream #0.1(eng): Audio: aac, 48000 Hz, stereo, s16, 58 kb/s At least one output file must be specified This is something you created from "original file", in a way that results in a broken container header. Can you, pretty please, give us the original mov file?
Command: ffmpeg -t 55 -i lord.mp4 -vcodec copy -acodec copy lord-half1.mp4 Command: ffmpeg -ss 55 -i lord.mp4 -vcodec copy -acodec copy lord-half2.mp4 Result: half1: Duration: 00:00:55.03, start: 0.000000, bitrate: 320 kb/s half2: Duration: 00:00:56.32, start: 0.000000, bitrate: 219 kb/s
Length mismatch. In the original file, there is a white blast between the scene with the Black Riders and Gandalf's face. It's missing in the end of half1 and beginning of half2.
I've rejoined the files to verify: # ffmpeg -i lord-half1.mp4 -vcodec mpeg2video -acodec copy lord-half1.mpg # ffmpeg -i lord-half2.mp4 -vcodec mpeg2video -acodec copy lord-half2.mpg # cat lord-half1.mpg lord-half2.mpg > lord-joined.mpg The missing scene is more visible now.
Normalization attempt: 1. 55'th second is 2637,25 frames (47.95 fps).
ffmpeg says the video stream is 47.90 fps, not 47.95 fps.
2. Rounding down to 2600 frames (gop 50) 3. 2600 frames is close to 54.223149113660062565172054223149 seconds 4. Rounding up to 54.2232 and crossing fingers 5. ffmpeg -t 54.2232 -i lord.mp4 gives us: Duration: 00:00:54.24 6. ffmpeg -ss 54.2232 -i lord.mp4 gives us file: Duration: 00:00:59.20 7. total length: 113.44 original length 111,27
a workaround attempt: 6. The -t 54.2232 gave us duration 54.24 so we will seek directly to it 7. ffmpeg -ss 54.24 -i lord.mp4 gives us file: Duration: 00:00:59.20 7. total length: 113.44 Looks clear!
Unfortunatelly the scene where Saruman throws Gandalf in the air is visible twice.
--
__ /.)\ +48 695 600 936 \(./ abodera at gmail.com _______________________________________________ ffmpeg-user mailing list ffmpeg-user at mplayerhq.hu https://lists.mplayerhq.hu/mailman/listinfo/ffmpeg-user
On Wed, Jan 6, 2010 at 3:27 PM, Jan Stary <hans at stare.cz> wrote:
This is something you created from "original file", in a way that results in a broken container header. Can you, pretty please, give us the original mov file?
http://www.ementor.pl/ffmpeg-split/lordoftherings_h640_ml.mov
Normalization attempt: 1. 55'th second is 2637,25 frames (47.95 fps). ffmpeg says the video stream is 47.90 fps, not 47.95 fps.
Recalculated, recoded... 1. 55'th second is 2634,5 frames (47.90 fps). 2. Rounding down to 2600 frames (gop 50) 3. 2600 frames is close to 54.279749478079331941544885177453 seconds 4. Rounding DOWN to 54.2797 5. ffmpeg -t 54.2797 -i lord.mp4 gives us: Duration: 00:00:54.32 6. ffmpeg -ss 54.2797 -i lord.mp4 gives us file: Duration: 00:00:59.20 7. total length: 113.52 original length 111,27 The halves overlap .... so no luck there. -- __ /.)\ +48 695 600 936 \(./ abodera at gmail.com
Don't know about timecode, but actually with 47.90 fps, a "chunk of length 00:00:10.0" doesn't even necessary correspond to one definite border frame. Right?
That's the whole point of timecode; it lets you simply number the frames so you can specify edit points, without worrying about how this relates to real time. Schemes like drop-frame timecode are designed to approximately correct these disparities (though obviously Michael Niedermeyer knows better than the entire film and TV industry, good thing we have him on board). That said none of this will help you with such an oddball frame rate. Where'd you get this file - is it double-rate 23.976 for some reason? There hasn't traditionally been drop-frame rates for 23.976. I'm not sure why this is, but I'd guess it's because feature films don't generally have to worry about a minute or so's duration either way. There are certainly drop-frame schemes for the other "doubled" rates that are becoming more prevalent, particularly in 720p HD. Anyway, this is all an aside to the original question. P
On Wed, Jan 6, 2010 at 4:15 PM, Phil Rhodes <phil_rhodes at rocketmail.com> wrote:
That said none of this will help you with such an oddball frame rate. Where'd you get this file - is it double-rate 23.976 for some reason? There hasn't traditionally been drop-frame rates for 23.976. I'm not sure why this is, but I'd guess it's because feature films [...]
Allright. Because ffmpeg doesnt have frame-accurate seeking we can only calculate gop and play with the proprietary hh:mm:ss.ms values of -ss and -t. As the frame rate has been an excuse from the beginning, I've recoded the file with: # ./ffmpeg -i ./lordoftherings_h640_ml.mov -vcodec libx264 -r 25 -g 50 -vpre default ./lord25.mp4 This means we now have: 25 fps 111.28 sec length keyframes every 2 sec. or 50 frames h264 But: Duration: 00:01:51.28, start: 0.000000, bitrate: 261 kb/s Stream #0.0(eng): Video: h264, yuv420p, 640x272, 196 kb/s, 25 fps, 25 tbr, 25 tbn, 50 tbc Stream #0.1(eng): Audio: aac, 48000 Hz, stereo, s16, 58 kb/s 50tbc ?? I'm clueless here.. also this: " Seems stream 0 codec frame rate differs from container frame rate: 50.00 (50/1) -> 25.00 (25/1)" Where did 50 come from? Let's split....... Half of the video is somewhere around 54 sec and it should be keyframe so let's do it: # ffmpeg -y -t 54 -i lord25.mp4 -acodec copy -vcodec copy lord25-half1.mp4 # ffmpeg -y -ss 54 -i lord25.mp4 -acodec copy -vcodec copy lord25-half2.mp4 And the results are in: 1. Duration: 00:00:56.28 (???) 2. Duration: 00:00:59.20 (that gives us a total of 115.48 seconds) I guess it means that -t 54 triggered jump to another keyframe. Let's try the last frame before 54 -t 53.9999: # ffmpeg -y -t 53.99999 -i lord25.mp4 -acodec copy -vcodec copy lord25-half1-1.mp4 Duration: 00:00:56.28 Still wrong. Let's try -t 53: # ffmpeg -y -t 53 -i lord25.mp4 -acodec copy -vcodec copy lord25-half1-2.mp4 Duration: 00:00:55.24 Wrong again However, when observing the output from encoder it states: frame= 1325 fps= 0 q=-1.0 Lsize= 2092kB time=55.17 bitrate= 310.6kbits/s video:1697kB audio:357kB global headers:0kB muxing overhead 1.840558% This means that the -t 53 correctly translates to 1325th frames (with 25fps). Any ideas? -- __ /.)\ +48 695 600 936 \(./ abodera at gmail.com
On Wed, Jan 6, 2010 at 16:41, Artur Bodera <abodera at gmail.com> wrote:
On Wed, Jan 6, 2010 at 4:15 PM, Phil Rhodes <phil_rhodes at rocketmail.com> wrote:
That said none of this will help you with such an oddball frame rate. Where'd you get this file - is it double-rate 23.976 for some reason? There hasn't traditionally been drop-frame rates for 23.976. I'm not sure why this is, but I'd guess it's because feature films [...]
Allright. Because ffmpeg doesnt have frame-accurate seeking we can only calculate gop and play with the proprietary hh:mm:ss.ms values of -ss and -t.
As the frame rate has been an excuse from the beginning, I've recoded the file with: # ?./ffmpeg -i ./lordoftherings_h640_ml.mov -vcodec libx264 -r 25 -g 50 -vpre default ./lord25.mp4
This means we now have: ? ?25 fps ? ?111.28 sec length ? ?keyframes every 2 sec. or 50 frames ? ?h264
But: ?Duration: 00:01:51.28, start: 0.000000, bitrate: 261 kb/s ? ?Stream #0.0(eng): Video: h264, yuv420p, 640x272, 196 kb/s, 25 fps, 25 tbr, 25 tbn, 50 tbc ? ?Stream #0.1(eng): Audio: aac, 48000 Hz, stereo, s16, 58 kb/s
50tbc ?? I'm clueless here.. also this: " Seems stream 0 codec frame rate differs from container frame rate: 50.00 (50/1) -> 25.00 (25/1)" Where did 50 come from?
Let's split....... Half of the video is somewhere around 54 sec and it should be keyframe so let's do it: # ffmpeg -y -t 54 -i lord25.mp4 -acodec copy -vcodec copy lord25-half1.mp4 # ffmpeg -y -ss 54 -i lord25.mp4 -acodec copy -vcodec copy lord25-half2.mp4
And the results are in: 1. Duration: 00:00:56.28 ?(???) 2. Duration: 00:00:59.20 ?(that gives us a total of 115.48 seconds)
I guess it means that -t 54 triggered jump to another keyframe. Let's try the last frame before 54 -t 53.9999: # ffmpeg -y -t 53.99999 -i lord25.mp4 -acodec copy -vcodec copy lord25-half1-1.mp4 Duration: 00:00:56.28 Still wrong.
Let's try -t 53: # ffmpeg -y -t 53 -i lord25.mp4 -acodec copy -vcodec copy lord25-half1-2.mp4 Duration: 00:00:55.24 Wrong again
However, when observing the output from encoder it states: frame= 1325 fps= ?0 q=-1.0 Lsize= ? ?2092kB time=55.17 bitrate= 310.6kbits/s video:1697kB audio:357kB global headers:0kB muxing overhead 1.840558%
This means that the -t 53 correctly translates to 1325th frames (with 25fps).
Any ideas?
Scene changes could be inserting keyframes automatically. Perhaps playing around with -sc_threshold you could generate more consistent GOPs. If you're re-encoding just to make cuts, and the re-encode is disposable, why not make it I-frames only? Then seeking would be far more accurate.
On Wed, Jan 6, 2010 at 11:33 PM, Dave P <lists.davep at gmail.com> wrote:
If you're re-encoding just to make cuts, [...]
I'm not. It's only for testing purposes. If I'm lucky I'll be able to probe the gop from the file I want to cut, but that's about it. Decoding to raw or iframes or frame images then encoding again is too expensive. I want to know how to cut it in half within existing stream (hence -acodec copy -vcodec copy). sc_threshold will probably affect the gop, but it will not help me if I don't know if scene changes. But logically -t x and -ss x should work.... -t x should crop video to nearest keyframe (to x), an -ss x should seek to nearest keyframe (to x). Regardles if it's 54, 55, 55.5 or 56th second. Why then am I getting overlapped chunks? -- __ /.)\ +48 695 600 936 \(./ abodera at gmail.com
Artur Bodera wrote:
On Wed, Jan 6, 2010 at 11:33 PM, Dave P <lists.davep at gmail.com> wrote:
If you're re-encoding just to make cuts, [...]
I'm not. It's only for testing purposes. If I'm lucky I'll be able to probe the gop from the file I want to cut, but that's about it. Decoding to raw or iframes or frame images then encoding again is too expensive. I want to know how to cut it in half within existing stream (hence -acodec copy -vcodec copy).
sc_threshold will probably affect the gop, but it will not help me if I don't know if scene changes.
But logically -t x and -ss x should work.... -t x should crop video to nearest keyframe (to x), an -ss x should seek to nearest keyframe (to x). Regardles if it's 54, 55, 55.5 or 56th second. Why then am I getting overlapped chunks?
I think you need to reed my former answer. -ss means movie-time not video-time
Hi ^^ I'm also seeking the way to split video accurately enough as possible. um, I think -ss and -t is not enough to do it. let's assume that if I have a 10:00 length video file, and I want to split it into two parts, one is of length of 5:00, and the other is of also 5:00; but if there're keyframes located at 4:55 and 5:05, ffmpeg must be trying to seek keyframe to split. I want to say that first part will be accurately 5:00, and second one will be 5:05 becuase of the fact that delta frames near the end border have their key frame, so there's no problem. but when second one trying to start at 5:00, it will recognize that there's no key frame to refer to start and make perfect frame, so it will ignore the -ss 00:05:00 option; I believe there's no way yet to avoid the duplication when splitting...only we can fix and customize the ffmpeg source;; and that's what I'm doing recently(that's too difficult to do...it's out of my ability (T.T)~~I just hope any brilliant developers help us~!) -- View this message in context: http://ffmpeg-users.933282.n4.nabble.com/Frame-gop-accurate-video-splitting-... Sent from the FFmpeg-users mailing list archive at Nabble.com.
On Jan 06 11:51:40, Artur Bodera wrote:
On Mon, Dec 28, 2009 at 12:48 PM, Artur Bodera <abodera at gmail.com> wrote:
[...] Chunks we want to have: ? 1)? ~30s (00:00:00 - 00:00:30) ? 2)? ~10s (00:00:30 - 00:00:40) ? 3)? ~71s (00:00:40 - 00:01:51)
File info: ? Duration: 00:01:51.27, start: 0.000000, bitrate: 256 kb/s ??? Stream #0.0(eng): Video: h264, yuv420p, 640x272, 192 kb/s, 47.90 fps, 23.98 tbr, 24k tbn, 47.95 tbc ??? Stream #0.1(eng): Audio: aac, 48000 Hz, stereo, s16, 58 kb/s
If I remember correctly, the file 'lord.mp4' was transcoded from a *.mov original as follows: ffmpeg -i ./lordoftherings_h640_ml.mov -vcodec libx264 -g 50 ./lord.mp4 Why do you transcode it at all? Doesn't this reduce the audio/video quality quite a bit? Does the same happen if you transcode with a bitrate higher then the cheap default? And if you leave out the "-g 50"? Anyway, does the same happen if you cut the MOV original into pieces instead of the (possibly crappy) MP4 transcode? Can you give us the original MOV file (01:51, can't be that big).
A small test - the easy way just using -ss and -t:
Encoding of chunk 1: ffmpeg -y -t 30 -i lord.mp4 -vcodec copy -acodec copy lord-c1.mp4 result: Duration: 00:00:30.05, start: 0.000000, bitrate: 286 kb/s Stream #0.0(eng): Video: h264, yuv420p, 640x272, 237 kb/s, 47.95 fps, 23.98 tbr, 48k tbn, 47.95 tbc Stream #0.1(eng): Audio: aac, 48000 Hz, stereo, s16, 43 kb/s
Encoding of chunk 2: ffmpeg -y -ss 30 -t 10 -i lord.mp4 -vcodec copy -acodec copy lord-c2.mp4 result: Duration: 00:00:15.27, start: 0.000000, bitrate: 226 kb/s Stream #0.0(eng): Video: h264, yuv420p, 640x272, 156 kb/s, 47.95 fps, 23.98 tbr, 48k tbn, 47.95 tbc Stream #0.1(eng): Audio: aac, 48000 Hz, stereo, s16, 64 kb/s
Encoding of chunk 3: ffmpeg -y -ss 40 -i lord.mp4 -vcodec copy -acodec copy lord-c3.mp4 result: Duration: 00:01:12.12, start: 0.000000, bitrate: 268 kb/s Stream #0.0(eng): Video: h264, yuv420p, 640x272, 199 kb/s, 47.95 fps, 23.98 tbr, 48k tbn, 47.95 tbc Stream #0.1(eng): Audio: aac, 48000 Hz, stereo, s16, 63 kb/s
After re-joining the video has 117.44 s (the original file had 111,27s). The fragments are inaccurate and overlap.
--
__ /.)\ +48 695 600 936 \(./ abodera at gmail.com _______________________________________________ ffmpeg-user mailing list ffmpeg-user at mplayerhq.hu https://lists.mplayerhq.hu/mailman/listinfo/ffmpeg-user
010/1/6 Jan Stary <hans at stare.cz>:
If I remember correctly, the file 'lord.mp4' was transcoded from a *.mov original as follows:
ffmpeg -i ./lordoftherings_h640_ml.mov -vcodec libx264 -g 50 ./lord.mp4
Why do you transcode it at all? Doesn't this reduce the audio/video quality quite a bit? Does the same happen if you transcode with a bitrate higher then the cheap default? And if you leave out the "-g 50"?
Well basically I'm looking for an universal way to calculate timecodes and cut any arbitrary length video. I just want to be able to cut any video in half or to more pieces in a way precise enough, so I can re-join the fragments later and still have the same video. The transcoding has been performed only for testing purposes - this way we know gop is exactly 50. It would be pointless to test -ss and -t with some exotic files encoded with god-knows-what and try to probe it with ffprobe. You can take any video file and transcode it with the same command line - then try to split it. If ffmpeg can't properly handle it's own files (encoded by ffmpeg) then I'm doing someting wrong or there's a bug. Nevertheless, here are the test files: lord.mp4 - http://www.ementor.pl/ffmpeg-split/lord.mp4 lord-c1.mp4 - http://www.ementor.pl/ffmpeg-split/lord-c1.mp4 lord-c2.mp4 - http://www.ementor.pl/ffmpeg-split/lord-c2.mp4 lord-c3.mp4 - http://www.ementor.pl/ffmpeg-split/lord-c3.mp4 -- -- __ /.)\ +48 695 600 936 \(./ abodera at gmail.com
On Jan 06 13:18:49, Artur Bodera wrote:
010/1/6 Jan Stary <hans at stare.cz>:
If I remember correctly, the file 'lord.mp4' was transcoded from a *.mov original as follows:
ffmpeg -i ./lordoftherings_h640_ml.mov -vcodec libx264 -g 50 ./lord.mp4
Why do you transcode it at all? Doesn't this reduce the audio/video quality quite a bit? Does the same happen if you transcode with a bitrate higher then the cheap default? And if you leave out the "-g 50"?
Well basically I'm looking for an universal way to calculate timecodes and cut any arbitrary length video. I just want to be able to cut any video in half or to more pieces in a way precise enough, so I can re-join the fragments later and still have the same video.
The transcoding has been performed only for testing purposes - this way we know gop is exactly 50. It would be pointless to test -ss and -t with some exotic files encoded with god-knows-what and try to probe it with ffprobe.
You can take any video file and transcode it with the same command line - then try to split it.
If ffmpeg can't properly handle it's own files (encoded by ffmpeg) then I'm doing someting wrong or there's a bug.
Nevertheless, here are the test files:
Can you give us the original MOV file?
lord.mp4 - http://www.ementor.pl/ffmpeg-split/lord.mp4
$ ftp http://www.ementor.pl/ffmpeg-split/lord.mp4 $ ffmpeg -i lord.mp4 FFmpeg version SVN-r13839, Copyright (c) 2000-2008 Fabrice Bellard, et al. configuration: --enable-shared --cc=cc --mandir=/usr/local/man --disable-debug --disable-optimizations --enable-gpl --enable-pthreads --enable-postproc --enab le-swscale --enable-liba52 --enable-libfaac --enable-libfaad --enable-libmp3lame --enable-libtheora --enable-libvorbis --enable-libx264 --enable-x11grab --extra -libs=-L/usr/local/lib -L/usr/X11R6/lib --extra-cflags=-I/usr/local/include -I/u sr/X11R6/include --disable-decoder=cavs --disable-altivec --disable-armv6 --disable-armvfp --disable-iwmmxt --disable-neon libavutil version: 49.7.0 libavcodec version: 51.57.2 libavformat version: 52.16.0 libavdevice version: 52.0.0 built on Jul 1 2009 12:49:39, gcc: 3.3.5 (propolice) [mov,mp4,m4a,3gp,3g2,mj2 @ 0x2fc58c70]edit list not starting at 0, ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ a/v desync might occur, patch welcome ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ Input #0, mov,mp4,m4a,3gp,3g2,mj2, from 'lord.mp4': Duration: 00:01:51.27, start: 0.000000, bitrate: 268 kb/s Stream #0.0(eng): Video: h264, yuv420p, 640x272, 23.98 tb(r) Stream #0.1(eng): Audio: libfaad, 48000 Hz, stereo Must supply at least one output file while your ffmpeg -i lord.mp4 says
File info: Duration: 00:01:51.27, start: 0.000000, bitrate: 256 kb/s Stream #0.0(eng): Video: h264, yuv420p, 640x272, 192 kb/s, 47.90 fps, 23.98 tbr, 24k tbn, 47.95 tbc Stream #0.1(eng): Audio: aac, 48000 Hz, stereo, s16, 58 kb/s
We might have different versions installed, but is it actually the same file? Something's wrong with the copy of lord.mp4 you gave us.
Also, On Jan 06 13:18:49, Artur Bodera wrote:
lord-c1.mp4 - http://www.ementor.pl/ffmpeg-split/lord-c1.mp4
$ ftp http://www.ementor.pl/ffmpeg-split/lord-c1.mp4 $ ffmpeg -i lord-c1.mp4 FFmpeg version SVN-r13839, Copyright (c) 2000-2008 Fabrice Bellard, et al. configuration: --enable-shared --cc=cc --mandir=/usr/local/man --disable-debug --disable-optimizations --enable-gpl --enable-pthreads --enable-postproc --enable-swscale --enable-liba52 --enable-libfaac --enable-libfaad --enable-libmp3lame --enable-libtheora --enable-libvorbis --enable-libx264 --enable-x11grab --extra-libs=-L/usr/local/lib -L/usr/X11R6/lib --extra-cflags=-I/usr/local/include -I/usr/X11R6/include --disable-decoder=cavs --disable-altivec --disable-armv6 --disable-armvfp --disable-iwmmxt --disable-neon libavutil version: 49.7.0 libavcodec version: 51.57.2 libavformat version: 52.16.0 libavdevice version: 52.0.0 built on Jul 1 2009 12:49:39, gcc: 3.3.5 (propolice) vvvvvvvvvvvvvvvvvvvvvvvvvvvvvv [mov,mp4,m4a,3gp,3g2,mj2 @ 0x2aac7c70]edit list not starting at 0, a/v desync might occur, patch welcome Seems stream 0 codec frame rate differs from container frame rate: 24000.00 (48000/2) -> 23.98 (24000/1001) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ Input #0, mov,mp4,m4a,3gp,3g2,mj2, from 'lord-c1.mp4': Duration: 00:00:30.05, start: 0.000000, bitrate: 286 kb/s Stream #0.0(eng): Video: h264, yuv420p, 640x272, 23.98 tb(r) Stream #0.1(eng): Audio: libfaad, 48000 Hz, stereo Must supply at least one output file which again differs from what you report. (Should I try with latest svn?)
On Wed, Jan 6, 2010 at 1:42 PM, Jan Stary <hans at stare.cz> wrote:
$ ffmpeg -i lord-c1.mp4 FFmpeg version SVN-r13839, Copyright (c) 2000-2008 Fabrice Bellard, et al. ?configuration: --enable-shared --cc=cc --mandir=/usr/local/man --disable-debug --disable-optimizations --enable-gpl --enable-pthreads --enable-postproc --enable-swscale --enable-liba52 --enable-libfaac --enable-libfaad --enable-libmp3lame --enable-libtheora --enable-libvorbis --enable-libx264 --enable-x11grab --extra-libs=-L/usr/local/lib -L/usr/X11R6/lib --extra-cflags=-I/usr/local/include -I/usr/X11R6/include --disable-decoder=cavs --disable-altivec --disable-armv6 --disable-armvfp --disable-iwmmxt --disable-neon ?libavutil version: 49.7.0 ?libavcodec version: 51.57.2 ?libavformat version: 52.16.0 ?libavdevice version: 52.0.0 ?built on Jul ?1 2009 12:49:39, gcc: 3.3.5 (propolice)
vvvvvvvvvvvvvvvvvvvvvvvvvvvvvv [mov,mp4,m4a,3gp,3g2,mj2 @ 0x2aac7c70]edit list not starting at 0, a/v desync might occur, patch welcome
Seems stream 0 codec frame rate differs from container frame rate: 24000.00 (48000/2) -> 23.98 (24000/1001) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Input #0, mov,mp4,m4a,3gp,3g2,mj2, from 'lord-c1.mp4': ?Duration: 00:00:30.05, start: 0.000000, bitrate: 286 kb/s ? ?Stream #0.0(eng): Video: h264, yuv420p, 640x272, 23.98 tb(r) ? ?Stream #0.1(eng): Audio: libfaad, 48000 Hz, stereo Must supply at least one output file
which again differs from what you report. (Should I try with latest svn?)
I don't see the difference: # ffmpeg -i lord-c1.mp4 FFmpeg version SVN-r20880, Copyright (c) 2000-2009 Fabrice Bellard, et al. built on Dec 22 2009 20:38:09 with gcc 4.3.2 20081105 (Red Hat 4.3.2-7) configuration: --prefix=/usr --incdir=/usr/include --libdir=/usr/lib --mandir=/usr/share/man --shlibdir=/usr/lib --enable-libfaac --enable-libfaad --enable-libgsm --enable-libmp3lame --enable-libtheora --enable-libvorbis --enable-libx264 --enable-libxvid --enable-x11grab --enable-avfilter --enable-postproc --enable-pthreads --enable-gpl --enable-debug --enable-avfilter --enable-avfilter-lavf --enable-nonfree libavutil 50. 7. 0 / 50. 7. 0 libavcodec 52.43. 0 / 52.43. 0 libavformat 52.43. 0 / 52.43. 0 libavdevice 52. 2. 0 / 52. 2. 0 libavfilter 1.12. 0 / 1.12. 0 libswscale 0. 7. 2 / 0. 7. 2 libpostproc 51. 2. 0 / 51. 2. 0 Seems stream 0 codec frame rate differs from container frame rate: 47.95 (48000/1001) -> 23.98 (24000/1001) Input #0, mov,mp4,m4a,3gp,3g2,mj2, from 'lord-c1.mp4': Metadata: major_brand : isom minor_version : 512 compatible_brands: isomiso2avc1mp41 encoder : Lavf52.43.0 Duration: 00:00:30.05, start: 0.000000, bitrate: 286 kb/s Stream #0.0(eng): Video: h264, yuv420p, 640x272, 237 kb/s, 47.95 fps, 23.98 tbr, 48k tbn, 47.95 tbc Stream #0.1(eng): Audio: aac, 48000 Hz, stereo, s16, 43 kb/s At least one output file must be specified -- __ /.)\ +48 695 600 936 \(./ abodera at gmail.com
participants (6)
-
abodera@gmail.com -
belcampo@zonnet.nl -
hans@stare.cz -
Innertia -
lists.davep@gmail.com -
phil_rhodes@rocketmail.com