Inconsistent Levels result with filter

[cornfluke]

Hello, long time listener, first time caller

So I am confused, I have a video shot in MP4 that is very dark, however I can see from my digital camera screen image information that can still be salvaged. Indeed playing in MPC with the libavcodec via ffdshow I can select levels, make a very steep level curve and still get some useable image out of the video. However when I turn around and try to make that change permanent via vdub I'm running into issues. for vdub I have the quicktime mp4 plugin as well as the ffdshow filters plugin, it loads in vdub, in filters I can go and select ffdshow and then levels and do a similar levels curve, but the results are much different, there is much less usable image resulting, and the bulk of the image becomes a lot more grey than I would expect, or than I saw via MPC.

I know there are some video things that ignore the far reaches of the input spectrum, say under 5 or over 235 or something (that I see in other menus in ffdshow), would something like this be the cause? Or is it just the way the quicktime plugin does it different.

Is there any way around this? I basically want to reencode my mp4 (prob. as xvid) but with the levels adjusted, so its just like I see it in MPC with ffdshow levels enabled, but I just can't figure a way.

What am I missing? (and free programs strongly preferred)
Thanks

[poisondeathray]

Are you judging this by vdub preview window or media player ?

Vdub can be set up differently than a media player (they might be using different renderers)

The other reason is vdub's RGB conversion when using filters. It converts Y'CbCr [16,235] => RGB [0,255] . Your "dark" video probably has most of the data <16 , but you lose the data <16, >235 before you even get to use a filter using vdub that way

You can "recover" the data if you adjust the levels in Y'CbCr colorspace (e.g. with avisynth), and you avoid RGB conversion. You can use levels() or smoothlevels()

[cornfluke]

thanks poisondeathray (again) yeah I was judging it via the preview window, but running it full through - same results.
and I bet they are running different (I don't think I could ever get the mp4 h.264 from ffdshow to work with vdub, so I used the quicktime plugin for it)

but I think you're right, which is what I suspected, that its clipping that range to start, it seemed like that might be the cause but I'm pretty ignorant of colorspaces, so I probably didn't pay enough attention to that.

I'll run it through avisynth and see what I get and repost, it might take me a bit since I haven't really used avisynth lately, and so never for h.264 video (newish camera)

[poisondeathray]

thanks poisondeathray (again) yeah I was judging it via the preview window, but running it full through - same results.
and I bet they are running different (I don't think I could ever get the mp4 h.264 from ffdshow to work with vdub, so I used the quicktime plugin for it)

If you load it with DirectShowSource() in avisynth, or use the directshow input driver for vdub , it should use ffdshow if you have it configured to decode your file . But for dark videos, I would still "fix" it in avisynth or before any RGB conversion

The decoder is different than the renderer. You can use the same decoder but the video will look entirely different if you use a different renderer. Try it out yourself. In MPC, try using vmr9 (renderless) , and switch to overlay mixer in the options. You may have to restart the media player between tests. Similarly, if you have directx enabled in the vdub configuration, then the graphics card overlay settings will govern the preview. If you disable it, the video can look different in the output pane preview (underlying video isn't affected)

[cornfluke]

awesome thanks, yeah I've thrown it via DirectShowsource, and so far it seems to be working, hurray. gonna tweak the numbers more and see what else I can get done with avisynth levels before going to vdub filters. I've tried to avoid any colorspace conversion so far in getting it out of that damn h.264, but never quite sure where its automatic

right, right, (the decoder/renderer difference I always see in gspot in the little test field at the bottom) but I never really think about it because until this issue, well... its never been an issue

i'm trying it out a couple ways, to see what steps I can skip between getting it into that tighter space via avisynth levels and when exactly I can convert to rgb cleanly so that I can use the other vdub filters, I suppose I could just do the colorspace conversion in avisynth as well, but I need to read up and understand colorspaces a little better first

thanks again, (and thanks for all the other threads I've used your advice from before when searching q's)

[jagabo]

The Brightness/Contrast filter in VirtualDub will allow you to access Y<16 if you use it first in the filter chain. Bump the brightness up by 7 or 8 percent. Then you can use Levels.

Or use ConvertToRGB(matrix="PC.601") in AviSynth to convert YUV 0-255 to RGB 0-255. If the video is high def use ConvertToRGB(matrix="PC.709") instead.