Sigmoid Module Math?

NateWeatherly · March 17, 2023, 7:13pm

I’d like to be able to replicate the tone curve aspect of the sigmoid module in high resolution 1D LUTs for use in DCP profiles and other editing apps but I’m struggling to figure out the math. I don’t really have any coding experience so I haven’t been able to dig into the code to figure it out yet. I’ve learned enough image processing math to be able to use Lattice (LUT creation/modification app with, I believe, python math expressions) to scale exposure and generate simple log-logistic curves (in the form of 1/(1+($x/.18)**-1.5) for example) but I can’t figure out exactly how the contrast and skew work or how to get the resulting curve to be centered around .18 rather than .5 so that it’s correct for linear data.

If anyone could point me in the right direction I’d appreciate it!

priort · March 17, 2023, 8:02pm

Taking a quick scan there is more going on than just a simple curve with a skew and contrast parameter changing the shape… at least from what I can see…

github.com

darktable-org/darktable/blob/master/src/iop/sigmoid.c

/*
    This file is part of darktable,
    Copyright (C) 2020-2022 darktable developers.

    darktable is free software: you can redistribute it and/or modify
    it under the terms of the GNU General Public License as published by
    the Free Software Foundation, either version 3 of the License, or
    (at your option) any later version.

    darktable is distributed in the hope that it will be useful,
    but WITHOUT ANY WARRANTY; without even the implied warranty of
    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
    GNU General Public License for more details.

    You should have received a copy of the GNU General Public License
    along with darktable.  If not, see <http://www.gnu.org/licenses/>.
*/
#ifdef HAVE_CONFIG_H
#include "config.h"
#endif

This file has been truncated. show original

NateWeatherly · March 17, 2023, 8:38pm

Wow, yeah, I guess there is! Do you know how I could see what’s happening in the “sigmoid_loglogistic_per_channel” program that it references? I’m assuming that’s where the actual math is?

kofa · March 17, 2023, 8:43pm

sigmoid_loglogistic_per_channel is the OpenCL kernel. Probably a plain C implementation would be easier to follow.
See darktable/src/iop/sigmoid.c at 4be1794967b781942a8bb3abf3b1cd331bddbed4 · darktable-org/darktable · GitHub and below.

Maybe the curve explorer will also be useful:

priort · March 17, 2023, 10:29pm

I thought this referred to one of the color preservations modes but I would defer to @kofa as I know he is a far better source than me… I thought at one time I remember a discussion that the curves are treated for each channel and then the ratio is maintained… that would be similar to filmic and why the ratio option looks more like filmic…again just going on memory here I could likely find it if I went digging and then it would be accurate and not a memory…

kofa · March 17, 2023, 10:57pm

There are C functions for the two colour modes:

per-channel mode:
– process_loglogistic_per_channel (if hue_preservation is practically disabled)
– process_loglogistic_per_channel_interpolated (if hue_preservation >= 0.001)
RGB ratio: process_loglogistic_rgb_ratio

And there are the OpenCL implementations: kernel_sigmoid_loglogistic_per_channel, kernel_sigmoid_loglogistic_per_channel_interpolated and kernel_sigmoid_loglogistic_rgb_ratio. Those live in sigmoid.cl.

jandren · March 18, 2023, 11:53am

Fun that you want use the curve outside of darktable!

You might have a easier time generating the curve from the Python implementation I did:

github.com

jandren/tone-curve-explorer/blob/master/tonecurves/loglogistic.py

import numpy as np

class LogLogistic(object):
    def __init__(self):
        self.scene_grey = 0.1845
        self.display_black = 0.0
        self.display_grey = 0.1845
        self.display_white = 1.0
        self.contrast = 2.01
        self.skew = -0.61
    
    def find_film_power(self):
        # Actual wanted slope should be the same as for without skew
        # and with constant display targets
        self.paper_power = 1.0
        self.film_power = self.contrast
        self.magnitude = 1.0
        self.fog = 0.0
        self.paper_e = pow(self.scene_grey, self.film_power) * ((self.magnitude / self.display_grey) - 1.0)
        target_slope = self.__calc_slope_at(self.scene_grey)

This file has been truncated. show original

Getting it to go through middle grey and having a orthogonal skew requires a bit of pre-computation of constants. Its much easier if you drop the skew part though.

Hope you manage to get it to work for your needs