What profiles does this now actually support?

[erazortt]

I think you guys (@Asd-g @quietvoid) have made some tests with sources other than profle 7, right?

[quietvoid]

Yes, only profile 5 still has desaturated colours compared to expected image.
I noticed the RGB to LMS coeffs aren't used, too.

Otherwise 8.4 is fine.

[erazortt]

In which curcumstances am I supposed to use the LSM coeffs? What processes LSM output?

[quietvoid]

If I refer to libplacebo:

// Colorspace transformation metadata
float nonlinear_offset[3];      // input offset ("ycc_to_rgb_offset")
struct pl_matrix3x3 nonlinear;  // before PQ, also called "ycc_to_rgb"
struct pl_matrix3x3 linear;     // after PQ, also called "rgb_to_lms"

I can't say if this being missing is the reason for the incorrect colours. It seems to affect chroma only.
Though as it's RGB, I guess it is because of a missing step.

It does PQ RGB > linearization > RGB to LMS matrix > delinearization (back to PQ).
So this seems to go right after the conversion to RGB.

[ValeZAA]

Yes, only profile 5 still has desaturated colours compared to expected image.

I do not get it, can profile 5 IPTPQc2 have actually FEL?? But even if it can, it does not matter, there is no content like that in the wild.

[graham8]

Yes, only profile 5 still has desaturated colours compared to expected image.

I do not get it, can profile 5 IPTPQc2 have actually FEL?? But even if it can, it does not matter, there is no content like that in the wild.

Profile 5 does not have FEL. It's single-layer.

@erazortt Can't this be closed, now?

[ValeZAA]

Erm, no, Profile 5 is 2 layers, it is just that 2nd layer is integrated as RPU into first layer. The queston is whether we can declare FEL in RPU. I doubt it.

[graham8]

Sorry. You are wrong. See the DV Profiles and Levels doc. FEL means there is video information in a second layer.

[ValeZAA]

EL means there is video information in a second layer.

Not always. 2nd layer may have no video information in the case of MEL. Or just have fake padding, zeroed out hevc.

the DV Profiles and Levels doc.

You mean dual layer specification from ETSI.

[graham8]

I said FEL means there is video infomation. MEL has no video information - agreed.

https://dolby.my.salesforce.com/sfc/p/#700000009YuG/a/4u000000l6G4/4R18riPaaW3gxpVx7XwyQLdEITLFjB.w.Si0LoQR5j8

[ValeZAA]

I said FEL means there is video infomation. MEL has no video information - agreed.

https://dolby.my.salesforce.com/sfc/p/#700000009YuG/a/4u000000l6G4/4R18riPaaW3gxpVx7XwyQLdEITLFjB.w.Si0LoQR5j8

No, FEL can be NLQ only haasn/libplacebo#286

Technically yes, NLQ is just first stage, you need video data ideally.
We need Haasn to answer what he meant.

[graham8]

So what you (or @hassn) are saying is that P5 can have (trivial) NLQ coefficients? I've never come across that. And surely it can safely be ignored. Trivial NLQs indicate MEL. Non-trivial NLQs would have nothing to do if there's no EL video.