Nikon HE* decoder

Processing
21

chroma seems solved, yet hardcoded and not derived from values in the headers, this will be the next step to look where to find the correction values in the file

row_mean_profiles
row_mean_profiles1668×949 204 KB

2 Likes
22

Current color rendering in darktable. Looks promising.

Missing:

  • Derive color correction from header or whitebalance metadata (still unclear)
  • Stone color / overall color seems still off to me
  • The Adobe DNG looks smoother so some deblocking is missing it seems
  • Decoding artifacts still present in the form of black/white dotted areas (all comps) and colored stripes (come from comp3)
  • Top looks smeared

grafik
grafik741×1069 403 KB

1 Like
23

Could anyone with access to a HE* camera and a color target shoot a neutral grey patch for me? Or best multiple color patches that include neutral grey as well? Thanks a lot! @martinus

Practical requirements for the shot:

  • HE* compression mode (obviously)
  • Flat, even lighting — avoid specular on the target surface
  • Keep the target reasonably large in frame so each patch spans several precincts
  • Shoot at base ISO to minimise noise in the Cr/Cb channels
  • Include a shot in HE (uncompressed or lossless) of the same target if your camera supports it — that would give us a direct before/after comparison of what the Nikon ICT does to the codec values, which is the most direct way to characterise the transform
2 Likes
24

I am currently on the road, could you give an visual example of what you are looking for?

25

something like https://www.xrite.com/de/categories/calibration-profiling/colorchecker-targets

26

I can see what I can do. As I am traveling right now I don’t have the option of getting them right now, unfortunately.

1 Like
27

now look at those beauties. I finally cracked the gains table. Banding is now completely gone

decoded_v127_premct_C0_Y
decoded_v127_premct_C0_Y4140×2760 6.63 MB

decoded_v127_premct_C1_Cr
decoded_v127_premct_C1_Cr4140×2760 6.05 MB

decoded_v127_premct_C2_Dg
decoded_v127_premct_C2_Dg4140×2760 5.22 MB

decoded_v127_premct_C3_Cb
decoded_v127_premct_C3_Cb4140×2760 7.76 MB

4 Likes
28

Awesome job!!

1 Like
29

I feel like Claude is my baby now, constantly telling it what not to do :smiley:

1 Like
30

Add it to the instructions so it “remembers”.

31

it’s a bit hard when you’re juggling a few free ones but I’ll manage :smiley:

32

Does anyone have an idea what they could have done to avoid this?

grafik
grafik757×817 72.5 KB

This is a 400% crop. The reference dng is much smoother and has more detail. Wavelet reconstruction seems fine so far so I must be missing something else. Or the wavelet reconstruction is still missing some formulas/indices I am not seeing.

Maybe some individual channels have too much / not enough intensity. So many degrees of freedom still unfortunately. Or the bayer grid is incorrectly assembled, or or…

33

I now have a calibration script for color. The more diverse images I have, the better. If you can’t upload them somewhere, let me know and I’ll make some room in my nextcloud