Hi
@user777!
Thanks again for your such detailed review and tests!
First of all, let me start with brief description of how image processed and AE algorithm works in chip used in this camera (please note, that is result of reverse engineering of firmware and experimenting with it, so that's only how I see it, and it could be absolutely wrong).
Raw image is obtained from sensor and some basic processing: defect pixels are replaced with neighbors' values interpolated, cropping and scaling done, noise reduction filter applied, debayering performed etc. At some point, in the middle of the processing, image has to be gamma-corrected, as old displays expected it in such form, and newer one just behave in the same way. At first glance, it seems to be trivial - just apply power function with 1/2.2, as display will reverse it with power function with 2.2, and should get the same data we have before gamma correction has been done. But in reality, different powers are used (1.8, 2.0, 2.4 etc), several standards exist (sRGB, rec.709 etc), and different formulas are used. So, curve could be built using that formula, and this curve is applied to RGB data, to each channel separately (or put it differently, non-linear digital gain/amplification is applied according to this curve). Internally this curve, or gamma, is stored as lut (or array) of 128 numbers. As input data has 10-bit resolution (range 0-1023), only 1/8 of points are in this lut, all missing points are interpolated. So, it is not very precise, but good enough to describe general form of curve. When all this processing is done and RGB data converted to YUV, luminance of image is calculated somehow, and this value is used by AE, which compares obtained luminance with target value and adjusts exposure and gain accordingly, also it has some additional parameters to adjust target based on current amount of light, overexposure detected etc. But the main point here is that gamma in this camera has influence on exposure, as absolutely same scene will have different luminance calculated with different gamma, and AE will work differently, so the scene will not be the same anymore
This is one of the trickiest things, in my opinion with this chip and it is very confusing. So, gamma could not be tweaked in the same way as in Photoshop, where e.g., shadows could be pushed down, and highlights up to increase contrast, here it may cause opposite result, as pushing shadows down can make luminance value too low, so AE adds more gain, which would lead to overexposure, and shadows go right into different segment of gamma for highlights, as result, overexposed image with low contrast and milky look and vice verse.
Now, let me explain what has been changed in last released mod and latest test mods. As you have noticed, difference in gamma, settings for AE, color and sharpness settings. Chip provides following functions to adjust color: Color Correction Matrix (CCM), Color Control (something similar to HSL, where Hue and Saturation could be adjusted globally, or across 24 sectors individually), and 3D Color Correction (3D lut with 9x9x9 dimensions, and is usually unused, as I can see in firmwares of this one and similar cameras). Coefficients used in CCM in original firmware are quite high, which results in strong colors and oversaturated image, instead of correcting sensor colors, and distort colors even more, or even clip data in some extreme cases. So, I have disabled it last mods for Git3 and Git2p, and emulated it with 3D LUT. Unfortunately, its resolution is too low, due to this fact and some additional restrictions I have applied during LUT calculation, produced colors were too weak, that is the reason, colorful or vivid effects were required to boost saturation. In latest test mods combined approach is used - CCM with lowest coefficients which are enough to get colors more or less correct and 3D LUT to selectively add saturation mostly in mid tones. As for me, it seems to be acceptable now, but LUT could be improved still.
Settings for sharpness actually are higher now in test mods, but thresholds and some other settings have been changed to make edge detection less sensitive and avoid sharpening of every single pixel with a bit different color. Latest test mod has Normal setting equal to Strong one from previous test mod (Normal from stock firmware), and Strong similar to Strong from stock (it could be reduced), and Soft disables sharpening in the same way, as in other mods. It is useful in some cases during experiments and in low light. I guess, disabling sharpening could be possible with B/W effect, but current purpose of this effect (same as Sepia) in mod, is to adjust video levels for players (VLC, most of players on Windows 7 etc) which treat it as Limited Range instead of Full Range, and clip partially shadows/highlights and show image with excessive contrast, while players which treat Limited Range as Full Range, will show it as low contrast and unsaturated.
Normal curve in latest test mod has only one change in highlights segment, which has been pushed down a bit to prevent soft clipping you have noticed previously on white objects in direct sunlight, and only top 20% of curve has been changed. Contrast increase mentioned with this test mod, is caused, probably, with some changes of AE settings, which are responsible for adjusting luminance target when overexposure is detected, and basically, just reduce exposure a bit in such case. Curve for High contrast now has midtone lifting applied you have suggested. As for me, its form looks too aggressive, but I must admit, it helps with overexposure in bright light, although I guess, it would cause more noise with high iso in low light due too high gain. But let's discuss curves in details a bit latter, as I need to render them as images.