Question regarding CRF and resolution

[teqnilogik]

I was reading up on how CRF works and I have a question regarding why a higher resolution encode can use a higher CRF value and maintain good quality.

http://mewiki.project357.com/wiki/X264_Settings#crf

According to the above page, it states that higher resolutions can get away with using higher CRF values. I don't understand why that is. The way I'm thinking is, the higher the resolution, the lower the CRF should be to maintain good quality since there is more detail. Why do higher resolution encodes look good with higher CRF values?

I realize the content of the video has a lot to do with how CRF encodes it, however, I'm just curious to why a higher resolution video can use a higher CRF value and get good results.

[nurbs]

Encoding artifacts are more noticable in low resolution encodes, especially when you upscale them to 720p or 1080p. At high resolutions the aritfacts are relatively small scale compared to the whole picture so you can get away with a higher CRF before they become noticable.

[Dark Shikari]

Because you're watching the high res and low res encodes on the same screen.

If you weren't upscaling the low res encodes, there'd be no difference. But in reality, whether it's a DVD or a Blu-ray, you're playing it on the same computer or TV.

[teqnilogik]

Thank you for the replies. That makes sense.

[creamyhorror]

The way I'm thinking is, the higher the resolution, the lower the CRF should be to maintain good quality since there is more detail.
This is a misconception. If CRF worked like that, then there'd be little reason to offer it as an alternative to bitrate-targeted 2pass.

The higher the resolution is, the higher the bitrate should be to maintain good quality. The CRF should remain the same, since CRF effectively targets a constant quality for every unit area of video.

This is assuming the video is played back unscaled. If during playback it's upscaled to fill the screen, then a CRF that is actually higher is acceptable for higher resolution sources since the degree of upscaling is lower.

[Keiyakusha]

Its not like my question worth new thread and actually somehow related to this topic, so...

The CRF should remain the same, since CRF effectively targets a constant quality for every unit area of video.

I often hearing that constant CRF is a constant quality so you even can do some tests, pick value you like and use it for everything. But I just made 3 encodes with exactly the same commandline, only applied denoising is different and got these results:
No denoise -> SSIM Mean Y:0.9923446 kb/s:1244.48
MVTools (light) -> SSIM Mean Y:0.9938718 kb/s:1132.00
DFTTEST (stronger) -> SSIM Mean Y:0.9953559 kb/s:1143.00

I can't tell the difference visually, especially that denoisers have very different effect on the video, but why SSIM differ so much? I expect it to be different maximum by 0.0005
I believe the same thing you can see with different resolutions or different sources.

[GodofaGap]

AFAIK x264 is not achieving to get the same SSIM when using CRF. Or the other way around: just because two videos have the same SSIM, does not mean they have the same quality.