Hi group!
On an exceptional weekend, instead of studying I went skiing. Long story short, four commands on the command line and about one hour (I did not time the process and I did not attend it) later:
<
http://vimeo.com/88614631>
The commands:
/home/yuv/bin/ffmpeg -i 00110.MTS -deinterlace -f mp4 -vcodec libx264 -crf 22 -preset ultrafast -s hd1080 -strict experimental -acodec aac -b 25000k -ab 160k -ar 44100 -ac 2 00110.mp4
transcode -J stabilize=shakiness=8:show=1,preview -i 00110.mp4 -y null,null -o dummy
transcode -J transform -i 00110.mp4 -o stab.avi -y raw
/home/yuv/bin/ffmpeg -i stab.avi -f mp4 -vcodec libx264 -crf 22 -preset slow -s hd720 -strict experimental -acodec aac -b:v 10000k -ab 160k -ar 44100 -ac 2 upload.mp4
The flow is really nice, especially compared with the shakiness of the original (the highest priority was not to bump into others while skiing backward), but it seems that during some sequences it placed bad control points (on the subject in motion), resulting in "subject-jitter". The options of the first transcode command visualize the selected control points.
The question is: given the sequence of individual frames, can cpfind do better at setting CPs only on the background and not on moving subjects? I suspect yes.
The corollary question is then: can it be integrated into transcode <
http://www.transcoding.org/> or into ffmpeg <
http://www.ffmpeg.org/>?
References:
* <
http://public.hronopik.de/vid.stab/>
* <
http://www.transcoding.org/transcode?Filter_Plugins/Filter_Stabilize>