Converting Fraps footage with x264, output file's colors become brighter?

[Dark Shikari]

Originally Posted by poisondeathray

If you're using the fraps decoder with AVISource, it will be RGB

Since when? FRAPS is a YV12 format.

[Blue_MiSfit]

To shrink PC range (0-255) to TV range, you can use ColorYUV(levels="pc->tv"), or better yet SmoothLevels(preset="pc2tv")

Derek

[cyberbeing]

Originally Posted by Dark Shikari

Since when? FRAPS is a YV12 format.

Since November 21st, 2009 I guess? In FRAPS 3.0.3 they added a Lossless RGB capture mode.

By default FRAPS still captures in their proprietary lossy YV12 compression, but by checking a box in setting, you can get proprietary lossless RGB compression instead, at the expense of slower capture speed and increased file size.

[sneaker_ger]

Also note that when using their decoder with AviSource() it seems to always decode to RGB32, even it's not RGB.

[Gfer]

I'm really confused now. What should I do to ensure the best compatibility? Convert to TV-range(Rec709)?

Edit: It seems like it was an issue with DXVA and my ATI drivers, using ffdshow or disabling dynamic contrast when using the internal DXVA filter in MPC-HC makes the video look a little better.

Edit: I'm pretty sure DXVA is messing up the colors. Looks just fine when using ffdshow for h264. Is there any fix for this? Disabling dynamic contrast makes it look a little better, but not quite as good as the original/preview. Both MPC-HC's internal DXVA filter and ffdshow's DXVA filter produces the same results. I could upload the video if anyone wants to see what I'm seeing.

[Blue_MiSfit]

1) Verify that without any conversion, you're getting RGB out of AVISource

2) If so, use ConvertToYV12(matrix="rec709"), and things should be totally fine

3) If not (you're getting some kind of YUV format), use histogram() to determine if the video is full or limited range

4) If it's limited range, encode as-is. If it's full range, either scale it with ColorYUV/SmoothLevels, or set --fullrange in x264.

Derek

[Gfer]

Encoded a version like that this time. When played with DXVA on, either ffdshow's or MPC-HC's internal filter, the colors become slightly brighter. When using ffdshow's h264 decoder everything looks fine. Even when using shaders in MPC-HC or setting a different output range in my graphic's driver, it still doesn't look like it should. Where do I go from here?

[Snowknight26]

Sounds like your GPU drivers are doing some post-processing.

Originally Posted by Dark Shikari

You must either convert to TV range in your script, or use --fullrange.

Most decoders ignore the fullrange flag, sadly. Even ffplay does, and ironically seems to do the PC -> TV conversion twice.

[Gfer]

Well, I have a ATI Radeon HD5850 GPU. There's Intelligent Saturation, Skin tone correction, dynamic range (which is 0-255) and dynamic contrast. Turning off these options and keeping the dynamic range at 0-255 made the video look like it should. Thanks for the help.

[Blue_MiSfit]

Indeed.

I prefer to turn this dynamic contrast /skin tone correction stuff off anyway - you shout set things up correctly without them.

My feeling on features like these is that they could potentially cause issues if you feed them unexpected input. I'm not basing that on anything but gut instinct, though !

Derek

[Snowknight26]

Now if only there was a way to keep the source input levels without Avisynth (such as when feeding x264 the input directly) and without --fullrange on..

[nm]

Originally Posted by Snowknight26

Now if only there was a way to keep the source input levels without Avisynth (such as when feeding x264 the input directly) and without --fullrange on..

Like this:

Originally Posted by Dark Shikari

We have a patch under development to make x264 automatically do --fullrange, but this will not work for Avisynth scripts (it's impossible to judge reliably from an Avisynth script whether the output is fullrange or not).

[Schrade]

Just to let you know, you want your video a bit brighter if you're going to upload it to YouTube. They seem to like to crush the blacks.

My current work process for FRAPS videos is using direct264 to encode the files and ffdshow to set the levels to 16-235. Direct264 reads the file in and it uses ffdshow to decode it on my system.

When encoding the videos using normal x264 and the --fullrange option, the videos playback locally perfectly fine but if you upload them to YouTube, YouTube will crush the blacks and make the videos really dark.

[Dark Shikari]

You should convert to TV range before uploading to Youtube.

[userix]

I have Fraps 3.2.9 and am using Handbrake to transcode my lossless RGB Fraps video captures. I am getting the "washed out" look with my transcoded videos. What arguments exactly do I have to set in my Handbrake CLI box to get it to convert with the right colorspace, luma, chroma, etc? I'm new to all this YUV, RGB, bt601, bt709 stuff.

How/where do I find and use this convertToYYV12 function? Thanks for the help!

[mzso]

Originally Posted by Dark Shikari

x264 by default encodes as TV range.

You must either convert to TV range in your script, or use --fullrange. We have a patch under development to make x264 automatically do --fullrange, but this will not work for Avisynth scripts (it's impossible to judge reliably from an Avisynth script whether the output is fullrange or not).

Ok. So at present, do I have to use --fullrange with ffmpeg?

How to verify the range with certainty? (The renderer might be good enough to guess/determine the range of the video)

[mzso]

Originally Posted by jq963152

There is another thread over there:

http://forum.doom9.org/showthread.php?t=155772

That seems like an avisynth topic which I don't use. Which part is of interest to me?

[vivan]

Originally Posted by mzso

Ok. So at present, do I have to use --fullrange with ffmpeg?

How to verify the range with certainty? (The renderer might be good enough to guess/determine the range of the video)

It's now --input-range PC in x264, probably the same in ffmpeg...

To verify it - open it in properly configured player, e.g. MPC-HC with LAV filters and madVR. madVR should show that range is full in OSD, and colors&brightness of the image should match source.

[mzso]

Originally Posted by vivan

It's now --input-range PC in x264, probably the same in ffmpeg...

To verify it - open it in properly configured player, e.g. MPC-HC with LAV filters and madVR. madVR should show that range is full in OSD, and colors&brightness of the image should match source.

I tried "-color_range 2"
Depending on decoder, I get full range (best guess), but with artifacts:



Otherwise limite range, and blurry. Same results without it.

Now, I'm going try to figure out how to pass --input-range PC to ffmpeg.

[mzso]

Originally Posted by jq963152

You asked this:





And the other thread has the answer to it, see:







By the way, the latest HandBrake version (0.9.9 was just released), converts FRAPS YV12 footage ("Force lossless RGB capture (may be slower)" checkbox disabled) better than anything else i'm aware of.

There is a thread about it:

http://forum.doom9.org/showthread.php?t=165680

With the "Force lossless RGB capture (may be slower)" checkbox enabled, HandBrake messes up the colors. But with that checkbox disabled (YV12), HandBrake is quite perfect at converting FRAPS YV12 to x264.

I really would like to know why HandBrake is so good at converting FRAPS YV12.

I still don't know how to convert FRAPS YV12 with AviSynth as good as HandBrake does.

But If I disable lossless "Force lossless RGB capture (may be slower)" I get a lossy video right?