Hi

I wonder whether you could help with my query, it relates to FFmpeg's reading of Canon's HV30 video and conversion of it to RGB. I take it that kdenlive works in RGB from the outset and not 'YUV' type sources.

The link below is to a screen grab of a project I was doing in Blender but I'm skeptical of what is happening to the video import and wonder as Kdenlive and Blender both use FFmpeg whether Kdenlive solves the issue I think is evident in the screen grab or not. And whether there is a problem at all. :-)

Although i mention blender here I'm not looking for comments about the applications VSE, it's interface or 'I don't know blender', the screenshot shows a Lumascope, vectorscope and histogram which I would imagine any keen hobbyist / Pro can read regardless of application. :-)

http://yellowsblog.files.wordpress.com/2009/09/screenshot.png

It shows my desktop, on the left screen is blenders import of a HV30 video shot in 25p mode and as comparison, the right screen shows the exact same footage, but first converted in AVISynth and exported as an RGB image sequence then imported into blender.

In AVISynth I used a ConvertToRGB(matrix="PC.709") to get the best full range conversion between YCrCb and RGB.

Blender uses FFMpeg to do that conversion to RGB. I believe that FFmpeg upto 0.5 is colour space agnostic and that the new libavfilter handles the colour space conversions from version 0.5.

The histogram on blenders import looks far more spiked/combed, the vectoscope more sparse and the luma scope levels are elevated / stretched, although black level looks much the same.

I haven't got to grips with kdenlive yet and ideally I should load the video in and test for myself. I'm not sure if kdenlive has scope tools to compare with.

So my query is does kdenlive do something special with FFmpeg or seperate code in kdenlive, perhaps you can spot immediately whats happening or a good guess in the right direction.

Thanks, looking forward to trying kdenlive out in preference to Blender.