search

pinterf / AviSynthPlus

AviSynth with improvements
http://avs-plus.net
208 stars 24 forks source link

Wrong UV planes when converting from Y32 #18

Closed magiblot closed 5 years ago

magiblot commented 6 years ago

Many conversion functions give weird results with 32-bit Y sources, most probably due to not being adapted to the new range for the U and V channels. It looks like they want to fill the U and V planes with half_range but they fill them with 0.5 instead of 0.0.

Y8 source clip: imatge

ConvertBits(32)
ConvertToYV12() #or YV16, or YV24, or RGB48, or RGB64, or PlanarRGB
ConvertBits(8)

imatge

ConvertBits(32)
ConvertToYV12() #or YV16, or YV24
Expr("x","x 0.5 -")
ConvertBits(8)

imatge

ConvertBits(16)
ConvertToYV12() #or YV16, or YV24
ConvertBits(8)

imatge

In addition, converting Y to YUY2 at any bitrate other than 8 will cause a severe crash:

imatge

I'm sorry.

pinterf commented 6 years ago

Thanks, chroma problem was fixed on git. And ConvertToYUY2 now gives error message when input bit depth is not 8, it was not guarded from hight bit depth input.

magiblot commented 6 years ago

Thank you very much!

Yeah, I had forgotten YUY2 is only intended for a bit depth of 8.

However, other Convert functions (YV12, YV16, YV24) keep the input bit depth, which is a bit confusing, considering the tables in http://avisynth.nl/index.php/Convert, although this may be the intended behaviour.

At first I thought they would downsample to output to 8 bit, but they seem to be synonims for ConvertToYUV420, YUV422 and YUV444, respectively.

I don't mean this should be changed, but I would find it very useful if it was a bit better explained in the Wiki page I linked (maybe it's already described in the changelog, but I haven't looked there).

pinterf commented 6 years ago

Yes, I have recognized that behaviour (ConvertToYV12..24 keeping the bitdepth), but forgot about it. It is not intended, perhaps it was introduced in an early internal reorganization. Maybe I'll look at it but it is not priority.

magiblot commented 6 years ago

Alright. Thanks!

magiblot commented 6 years ago

By the way, I have noticed when converting from RGB to YUV that RGB->YV24->YV12 is notably faster than RGB->YV12. Do you know why? My processor supports AVX at most.

pinterf commented 6 years ago

RGB is always converted to YV24 internally before doing YV12 conversion. So there is no reason to get noticable difference. You may start a different ticket on that and I'll check it.

magiblot commented 6 years ago

Okay. I'll do that.