Using Deshaker plugin for VirtualDub on CFHD at DVinfo.net

Earl Laamanen · April 3rd, 2010, 10:25 AM

I'm using the Deshaker v2.4 plugin for VirtualDub to steady some mountain bike video from a ContourHD camera, it's converted first using NeoHD to medium quality CFHD, then passed through VDub for deshaking only.

During the second pass, after the corrections are calculated, I'm reprocessing video data using CFHD codec v6.0, medium quality.

While I expect some picture quality loss due to resampling of rotated frames, I'm noticing some major picture/colour quality loss, especially a LOT of loss in dark areas, where there used to be enough information to pull out some detail with First Light gamma correction, it looks like the whole bottom end is lost.

Am I doing something wrong? Is there some reason to expect this when re-encoding CFHD medium to CFHD medium?

Will First Light corrections on the source video be burned into the output video? I guess it should but even that wouldn't explain the severe quality loss. I'll post some examples after I'm done processing the current video.

David Newman · April 3rd, 2010, 10:57 AM

My guess, it is deshaker / virtualdub is losing the shadow detail, but through no fault of their own. These tools are likely using computer graphics RGB 0-255. Yet your source and the CineForm conversions are YUV 16-235 (mostly) but some details will end up in the 0-15 and 236-255 ranges -- and FirsLight can no problem using this data unless you chop it off. When YUV is converted to RGB only the middle 16-235 is used (the rest are clipped.)

Solutions:

1) Run you whole tool chain in YUV, if Deshaker supports that.

2) Use FirstLight to squeeze your output range so that RGB 0-255 can hold all the source information. Basically this is a cgRGB to vsRGB (video systems) conversion. Set Lift to 0.06 and Gain 0.86. Do no other corrections, go through Deshaker, now you haven't lost anything.

OR

3) The easiest. Set the decode to default to Video System RGB. Import any CineForm clip into VirtualDub. Go straight to setting you video compression, and select CineForm, and click on "Configure". Set Video Systems for encode and decode.

Earl Laamanen · April 3rd, 2010, 01:30 PM

Thanks, I thought that might be the case but I wasn't sure where to correct for it. I guess I need to read up a bit on YUV and find the best pipeline. I used your third suggestion, it seems to work well. Everything I do with video to date is intended for PC playback anyway.

April 3rd, 2010, 10:25 AM	#1
Earl Laamanen New Boot Join Date: Dec 2008 Location: Prague, CZ Posts: 14	Using Deshaker plugin for VirtualDub on CFHD I'm using the Deshaker v2.4 plugin for VirtualDub to steady some mountain bike video from a ContourHD camera, it's converted first using NeoHD to medium quality CFHD, then passed through VDub for deshaking only. During the second pass, after the corrections are calculated, I'm reprocessing video data using CFHD codec v6.0, medium quality. While I expect some picture quality loss due to resampling of rotated frames, I'm noticing some major picture/colour quality loss, especially a LOT of loss in dark areas, where there used to be enough information to pull out some detail with First Light gamma correction, it looks like the whole bottom end is lost. Am I doing something wrong? Is there some reason to expect this when re-encoding CFHD medium to CFHD medium? Will First Light corrections on the source video be burned into the output video? I guess it should but even that wouldn't explain the severe quality loss. I'll post some examples after I'm done processing the current video.

April 3rd, 2010, 10:57 AM	#2
David Newman CTO, CineForm Inc. Join Date: Jul 2003 Location: Cardiff-by-the-Sea, California Posts: 8,095	My guess, it is deshaker / virtualdub is losing the shadow detail, but through no fault of their own. These tools are likely using computer graphics RGB 0-255. Yet your source and the CineForm conversions are YUV 16-235 (mostly) but some details will end up in the 0-15 and 236-255 ranges -- and FirsLight can no problem using this data unless you chop it off. When YUV is converted to RGB only the middle 16-235 is used (the rest are clipped.) Solutions: 1) Run you whole tool chain in YUV, if Deshaker supports that. 2) Use FirstLight to squeeze your output range so that RGB 0-255 can hold all the source information. Basically this is a cgRGB to vsRGB (video systems) conversion. Set Lift to 0.06 and Gain 0.86. Do no other corrections, go through Deshaker, now you haven't lost anything. OR 3) The easiest. Set the decode to default to Video System RGB. Import any CineForm clip into VirtualDub. Go straight to setting you video compression, and select CineForm, and click on "Configure". Set Video Systems for encode and decode. Attached Thumbnails __________________ David Newman -- web: www.gopro.com blog: cineform.blogspot.com -- twitter: twitter.com/David_Newman

April 3rd, 2010, 01:30 PM	#3
Earl Laamanen New Boot Join Date: Dec 2008 Location: Prague, CZ Posts: 14	Thanks, I thought that might be the case but I wasn't sure where to correct for it. I guess I need to read up a bit on YUV and find the best pipeline. I used your third suggestion, it seems to work well. Everything I do with video to date is intended for PC playback anyway.