Yellow_
25th April 2010, 15:12
Hi
New camera, new agro. :-) h264 AVC in Quicktime, Canon T2i/550D
I need to use h264 AVC video sources with a compositor that only handles RGB. Last camera I was doing this with was HDV mpeg2 and successfully did it without luma scaling using VDubMod, AVISynth ConvertToRGB and PC.709, save as image seq or uncompressed AVI.
I'm doing the same process now for the AVC QT's using Apples latest version of QT on Win XP. I'm using QTSource.dll (v. 0.0.8 from the authors site dated about 2007) and convert with PC.709.
I export as .png image seq's via Vdub 1.9.8, a few frames at first as a check.
I check them in ImageJ for the numeric histogram list it gives and The Gimp for it's histogram and Colour Cube Analysis.
I don't really consider this pixel peeping, I'm trying to maintain the quality of the 8bit sources with emphasis on grading later, not playback.
Here's the problem:
I'm getting scaled luma, evident by the clear, equally spaced gaps across the histogram in the Gimp. But the ImageJ histogram numeric list shows values from 10 to 255, suggesting there has been no luma scaling.
But doing a colour cube analysis in The Gimp on the png shows I have about 60000 unique colours, typical of the sort of values from using FFMPEG. Which I refuse to put near my camera sources.
Now if I do the same process using Avidemux, which reads the QT AVC in and I save it as YV12 Raw AVI and do the ConvertToRGB AVIsynth process I get no scaling, no gaps in the histogram and about 130000 unique colours, double that of the QTSource route.
This is not unusual, my HDV sources converted 'correctly' ie no scaling would yield 130000 ish unique colours per 1920x1080 frame .png. If I put FFMPEG anywhere near this process even for YUV->YUV conversions I'd loose half the unique colours.
So back to QTSource problem, what I think is happening is a 16 - 235 to 0 - 255 luma scaling and then a second luma scaling back to YV12 sort of levels. Giving me gaps in the histogram that shouldn't be there.
Does QTSource (QTInput) do anything here? Or just get the frames from QT. Is it getting them via the Player? Could that do a YV12 to RGB conversion before AVISynth gets it?
Other question, I can't find much info on Avidemux YV12 RAW format, should I use this and forget the whole QTSource route? Bearing in mind I'm looking for the best quality conversion I can having to go from YV12 to RGB.
Also why does FFMPEG make what appears to be such a poor job of YUV->YUV conversions with regard to reduced unique colours? Giving half what I get avoiding it. :-)
I'm a noob to all this so I may well have totally got it wrong. Feel free to tell me the Gimps Colour Cube Analysis is pointless, that I'm wasting my time. Gotta learn some how. :-)
New camera, new agro. :-) h264 AVC in Quicktime, Canon T2i/550D
I need to use h264 AVC video sources with a compositor that only handles RGB. Last camera I was doing this with was HDV mpeg2 and successfully did it without luma scaling using VDubMod, AVISynth ConvertToRGB and PC.709, save as image seq or uncompressed AVI.
I'm doing the same process now for the AVC QT's using Apples latest version of QT on Win XP. I'm using QTSource.dll (v. 0.0.8 from the authors site dated about 2007) and convert with PC.709.
I export as .png image seq's via Vdub 1.9.8, a few frames at first as a check.
I check them in ImageJ for the numeric histogram list it gives and The Gimp for it's histogram and Colour Cube Analysis.
I don't really consider this pixel peeping, I'm trying to maintain the quality of the 8bit sources with emphasis on grading later, not playback.
Here's the problem:
I'm getting scaled luma, evident by the clear, equally spaced gaps across the histogram in the Gimp. But the ImageJ histogram numeric list shows values from 10 to 255, suggesting there has been no luma scaling.
But doing a colour cube analysis in The Gimp on the png shows I have about 60000 unique colours, typical of the sort of values from using FFMPEG. Which I refuse to put near my camera sources.
Now if I do the same process using Avidemux, which reads the QT AVC in and I save it as YV12 Raw AVI and do the ConvertToRGB AVIsynth process I get no scaling, no gaps in the histogram and about 130000 unique colours, double that of the QTSource route.
This is not unusual, my HDV sources converted 'correctly' ie no scaling would yield 130000 ish unique colours per 1920x1080 frame .png. If I put FFMPEG anywhere near this process even for YUV->YUV conversions I'd loose half the unique colours.
So back to QTSource problem, what I think is happening is a 16 - 235 to 0 - 255 luma scaling and then a second luma scaling back to YV12 sort of levels. Giving me gaps in the histogram that shouldn't be there.
Does QTSource (QTInput) do anything here? Or just get the frames from QT. Is it getting them via the Player? Could that do a YV12 to RGB conversion before AVISynth gets it?
Other question, I can't find much info on Avidemux YV12 RAW format, should I use this and forget the whole QTSource route? Bearing in mind I'm looking for the best quality conversion I can having to go from YV12 to RGB.
Also why does FFMPEG make what appears to be such a poor job of YUV->YUV conversions with regard to reduced unique colours? Giving half what I get avoiding it. :-)
I'm a noob to all this so I may well have totally got it wrong. Feel free to tell me the Gimps Colour Cube Analysis is pointless, that I'm wasting my time. Gotta learn some how. :-)