search

mixxxdj / mixxx

Mixxx is Free DJ software that gives you everything you need to perform live mixes.
http://mixxx.org
Other
4.25k stars 1.24k forks source link

New GLSL waveforms are less detailed than legacy waveforms #12641

Open fwcd opened 5 months ago

fwcd commented 5 months ago

Bug Description

Moving https://github.com/mixxxdj/mixxx/issues/12448#issuecomment-1874799386 to a new issue here. Comparing the current GLSL waveforms:

Screenshot 2024-01-24 at 18 04 29

to the legacy GLSL waveforms:

Screenshot 2024-01-24 at 18 04 38

...there is a fair bit of detail missing in the new GLSL waveforms, especially the bass-heavy parts display more structure.

Version

2.5-alpha-207-g7ab184441d (HEAD)

OS

macOS 14.3

mxmilkiib commented 5 months ago

Looks mono, not stereo

Edit: ah I see #11837

fwcd commented 5 months ago

Here's another example of GLSL

Screenshot 2024-02-01 at 16 29 36

vs. GLSL (legacy)

Screenshot 2024-02-01 at 16 29 44

I've found myself switching back to the legacy waveforms for live use, the bass peaks are just much more recognizable.

ronso0 commented 5 months ago

Related https://github.com/mixxxdj/mixxx/issues/11833#issuecomment-1905165836

How shall we proceed here? Regardless the huge performance improvement, is this a regression? Is it severe enough to be fixed before the release?

daschuer commented 5 months ago

Since the legacy waformsvate still there, this should not block the release.

mxmilkiib commented 5 months ago

Are these top examples still with the curve that makes it look clipped when it's not? That's the only reason I can imagine it looks that inflated. It's like it lost a battle in the normalisation wars. Or is that new example current 2.5-alpha?

Re colour, in the other issue, the fact that green never appears to be used is buggy to me.

A skin with only waveforms n library would be good for comparisons.

fwcd commented 5 months ago

Both screenshots were taken with the (then) most recent 2.5/main builds

mxmilkiib commented 5 months ago

I don't get it then, it literally looks clipped. That boost curve was removed, iirc, yes? If so, why then does it look normalised in the first example and clipped in the second? Edit: there's also a lot of headroom also, so that's confusing me too, but that could just be the track(?).

acolombier commented 1 month ago

It looks like this is impacting all waveforms with various degree of difference. Here is a video which might help to illustrate better these difference.

https://github.com/mixxxdj/mixxx/assets/7086688/45c2a2b0-97c4-4504-aa9a-0cbcc2c45e12

(Only the beginning is relevant, the rest is demo for PR #13220)

ywwg commented 1 month ago

degree of difference

have you tested with the texture PR?

acolombier commented 1 month ago

I haven't no, will try and merge the branch in mine and share a video when I can

acolombier commented 1 month ago

Here is a comparison with RGB and Filtered and it looks even more different (except the color, which is fixed for filtered)

https://github.com/mixxxdj/mixxx/assets/7086688/a24813d4-78e0-4977-8a8d-8d0a8938a622

ywwg commented 1 month ago

we really need to fix the gain issues with our waveforms, it makes direct comparison impossible. Either 2x everywhere or nowhere, not in some of them