Recently in FFmpeg Category

Yesterday a new version of popular Linux video editing tool, Kino, was released. The new version is 1.3.0 and contains the following changes:

• Updated export scripts for FFmpeg changes (x264, mp3)
• Improved speed on SMP systems by enabling FFmpeg multi-threaded codecs
• Improved import (DV conversion) progress dialog
• Added gstreamer-based Ogg Theora to the blip.tv publishing script
• Added quality level option to the blip.tv publishing script
• Updated Hungarian translation
• Added Ukranian translation by Yuri Chornoivan

Congratulations to Dan Kennedy and the team.

The new source files can be downloaded directly from here.

Unfortunately my Linux based non-linear editing tool of choice, Open Movie Editor, doesn't currently support directly altering video playback speed. For example, if you wanted a portion of your new compilation to run at 200% of original recorded speed, it can't be done within OME. This exact functionality was something I needed for an existing editing project.

After some thought and investigation, such changes can be achieved through using a combination of FFmpeg and yuvfps, which is part of mjpeg tools, to alter the framerate of the desired footage. If your original file is PAL based, with a framerate of 25fps, changing the framerate to 50fps will result in the video running twice as fast, for half as long.

I didn't initially have mjpegtools installed, but on my Debian based system this was easy enough with

sudo apt-get install mjpegtools

Next, the input video needs to be converted to yuv4mpegpipe format, passed through yuvfps and output to a new avi file. Here's the command line I used to create a clip at 50fps:

ffmpeg -i input.dv -f yuv4mpegpipe - | yuvfps -s 50:1
-r 50:1  | ffmpeg -f yuv4mpegpipe -i - -b 28800k -y output.avi

Change the 50:1 ratios to whatever framerate you require. e.g. 100:1 for 100fps. Be sure to set the output file bitrate to a relevant quality level. Omitting this flag will result in a poor quality AVI output file by default.

The resulting AVI file was easily played back with Totem, and handled on the timeline admirably by OME.

Thanks to Victor Paesa on the FFmpeg mailing list for pointing me in the right direction.

Some other options to investigate include the new Libavfilter for FFmpeg and converting the original footage to a raw data file, which will lost the audio.

Extracting all frames from a video file is easily achieved with FFmpeg.

Here's a simple command line that will create 25 PNG images from every second of footage in the input DV file. The images will be saved in the current directory.

ffmpeg -i input.dv -r 25 -f image2 images%05d.png

The newly created files will all start with the word "images" and be numbered consecutively, including five pre-appended zeros. e.g. images000001.png.

From a video that was 104 seconds long, for a random example, this command would create 2600 PNG files! Quite messy in the current directory, so instead use this command to save the files in a sub-directory called extracted_images:

ffmpeg -i input.dv -r 25 -f image2 extracted_images/images%05d.png

Moving on, let's say you just wanted 25 frames from the first 1 second, then this line will work:

ffmpeg -i input.dv -r 25 -t 00:00:01 -f image2 images%05d.png

The -t flag in FFmpeg specifies the length of time to transcode. This can either be in whole seconds or hh:mm:ss format.

Making things a little more complex we can create images from all frames, beginning at the tenth second, and continuing for 5 seconds, with this line:

ffmpeg -i input.dv -r 25 -ss 00:00:10 -t 00:00:05 -f image2 images%05d.png

The -ss flag is used to denote start position, again in whole seconds or hh:mm:ss format.

Maybe extracting an image from every single frame in a video, resulting in a large number of output files, is not what you need. Here's how to create a single indicative poster frame, of the video clip, from the first second of footage:

ffmpeg -i input.dv -r 1  -t 00:00:01 -f image2 images%05d.png

Notice that the -r flag is now set to 1.

If you want the poster frame from a different part of the clip, then specify which second to take it from using the -ss tag, in conjunction with the line above.

Lastly, if you wanted to create a thumbnail story board, showing action throughout the entire length of the video clip, you'll need to specify the output image dimensions. Use the following line:

ffmpeg -i input.dv -r 1 -f image2 -s 120x96 images%05d.png

My original file was 720x576, so the image dimensions are a whole division of this.

A request on the Ubuntu forums asked for some assistance creating x264 files from footage originating on DVD. The following FFmpeg command represents input from a couple of users regarding what might be the best options:

ffmpeg -y -i input_file -an -v 1 -threads auto -vcodec libx264 -deinterlace -b 5000k -bt 175k -flags +loop -coder ac -refs 1 -loop 1 -deblockalpha 0 -deblockbeta 0 -parti4x4 1 -partp8x8 1 -me epzs -subq 1 -me_range 21 -chroma 1 -slice 2 -bf 3 -b_strategy 1 -level 30 -g 300 -keyint_min 30 -sc_threshold 40 -rc_eq 'blurCplx^(1-qComp)' -qcomp 0.7 -qmax 51 -qdiff 4 -i_qfactor 0.71428572 -maxrate 5000k -bufsize 2M -cmp 1 -s 720x480 -f mp4 -pass 1 /dev/null

ffmpeg -y -i input_file -v 1 -threads auto -vcodec libx264 -deinterlace -b 5000k -bt 175k -flags +loop -coder ac -refs 5 -loop 1 -deblockalpha 0 -deblockbeta 0 -parti4x4 1 -partp8x8 1 -me full -subq 6 -me_range 21 -chroma 1 -slice 2 -bf 3 -b_strategy 1 -level 30 -g 300 -keyint_min 30 -sc_threshold 40 -rc_eq 'blurCplx^(1-qComp)' -qcomp 0.7 -qmax 51 -qdiff 4 -i_qfactor 0.71428572 -maxrate 5000k -bufsize 2M -cmp 1 -s 720x480 -acodec libfaac -ab 256k -ar 48000 -ac 2 -f mp4 -pass 2 new_file.mp4

This command, as an overview does the following:

• Uses libx264 as the output video codec
• Uses libfaac as the output audio codec
• Deinterlaces the original DVD sourced footage
• Allows FFmpeg to choose the number of threads to use for multi-core systems
• Sets the output video bitrate at 5000kbps (or roughly 5Mbps)
• Sets the output audio bitrate at 256kbps
  
• Deblocks the output footage
• Uses CABAC encoding
• Uses .mp4 as the output file container
• Uses B-Frames
• Uses 2 pass encoding - directing the first output to /dev/null and the second pass to a new file

There are of course many other options included in this command. Further useful reading can be found on the FFmpeg documentation page:

http://ffmpeg.mplayerhq.hu/ffmpeg-doc.html

Also, this Mencoder specific page has some useful information regarding encoding using x264:

http://www.mplayerhq.hu/DOCS/HTML/en/menc-feat-x264.html

Recently I took the time to run a quick comparison test using FFmpeg for producing content with the x264 and xVid codecs. x264 produces H.264 content, while xVid encodes in MPEG4. I was a little surprised with the results.

The input file was sourced from BBC Motion Gallery. It was an MPEG2 Program Stream with I-Frames, encoded at approximately 50Mbps. It also contained a single MP2 audio track at approximately 356kbps. To see the clip on BBC Motion Gallery, click here. The clip was chosen because it is only 2 seconds long, so I could transcode it quickly, and there is also lots of movement, so I expected artifacts.

The output file container is QuickTime MOV, the video bitrate around 2Mbps, the audio codec aac (through libfaac) at 128kbps. I performed a 1 pass and 2 pass encode with x264 and xVid. The output file was to be 720x404 in size, this is 16:9 aspect ratio. All files were played back on a Windows XP machine using QuickTime Player 7.1.3. Some cropping of the original MPEG2 was required to remove VITC at the top and some black padding left and right. An example FFmpeg command line I used is outlined below. As you can see, there are very few optimisations other than the base settings.

FFMPEG command for the second pass x264 encode:

ffmpeg -i 2573-9.mpg -vcodec libx264 -f mov -b 2000k -acodec libfaac -ab 128k -s 736x442 -croptop 34 -cropbottom 4 -cropleft 8 -cropright 8 -deinterlace -pass 2 2573-9_h264.mov

The final output file sizes are as follows:

• x264 1 pass: 599.558 kilobytes
• x264 2 pass: 553.767 kilobytes
• xVid 1 pass: 577.232 kilobytes
• xVid 2 pass: 559.947 kilobytes

So, while xVid one pass produced smaller file sizes than x264 one pass, the x264 2 pass file is smaller than the xVid 2 pass file. Due to the short duration of the input file, no comparison of encoder speed could really be made. However, anecdotally from other ad hoc encoding jobs, xVid does seem to be quicker.

Now, to the real proof of the pudding, what was the quality like. Have a look at the following image file, click the thumbnail for a larger version:

This is where I was surprised. The x264 files are on the left. 1 pass is bottom left. 2 pass is top left. The xVid files are on the right. 1 pass bottom right. 2 pass top right. Clearly, the 1 pass xVid file is superior to the 1 pass x264 file. I also believe that the 2 pass xVid file is slightly better quality then the 2 pass x264 file. Areas for close inspection (we're looking at the two top files here):

• Bottom right corner. There is more blocking and artifacts on the x264 file.
• Top right wing tip. The xVid file has better definition here.
• The underside of the wings and tail. Again the xVid file has better definition.
• The background fire. I think the xVid file has less artifacts and better definition in general.
• As there aren't a lot of colours in the example video, it's hard to say which codec handles colour better, but there is obviously more depth in each of the 2 pass samples, compared to the 1 pass output.

I was surprised that to my eye the xVid content appeared to be superior to the x264 output. Perhaps with a more complext FFmpeg command and options this would have changed.