[TEST] IR Dust Removal

[webmogul1]

100% Crops

[marcinz606]

can you share the raw for testing? i think some tweaks could be done to inpainting in high contrast areas like those letters

[webmogul1]

I think you have it already. It is the one with the sign in the middle.
Or did you delete it?

[thetalkingdrum]

Interesting. Openice fine mode seems to smother the grain way more than I can remember NikonScan doing on «Fine». But maybe my memory is off.

I also made the discovery recently that NikonScan seems to use different thresholds when scanning in negative and positive mode.
When scanning in positive mode fine seems to roughly equal normal in negative mode. I should do some tests and share the results with both open source ice projects.

[webmogul1]

Just tried the same negative with Negpy v0.46.
Negpy looks better now with the added grain.

But included openIce looks worse than version 1.0.1 from repository. It looks like grain is not being added??

No Ice

Negpy 0.46

Included OpenIce

OpenIce 1.0.1

[marcinz606]

included openice is basically dirty python port, not 1:1 binary application

[webmogul1]

included openice is basically dirty python port, not 1:1 binary application

Apparently, Claude removed it because it was redundant and slowed the process.
I don't know if it would slow the process, but it sure looks better with the extra(?) dither.

from openice.py:
Two departures from the original: no dither (the ℓ=3 band already restores real grain, and
it is the only serial dependency in the algorithm), and the level constants are measured
per frame rather than fixed — see calibrate.

and from here: #724
Dropped the LCG dither (the finest band already restores real grain, and it was the only serial dependency in an otherwise per-pixel algorithm) and added a per-component route cap — the film rebate sits above the dead floor on some scanners and arrived as one component covering 1.2% of a flatbed frame, blowing the inpaint budget and taking the real dust with it.

Maybe @a6o have something to say about that?

[a6o]

I've not read NegPy code, but ℓ=3 band is restoring pixel values from the same pixel. This could restore real grain, but if that pixel is fully covered by dust (which happens more often than not), then ICE recovers the value only from the surrounding pixels, leading to no grain.
I think that is what is happening in the attached image above. I would say dither is critical part of the Digital ICE program.

[webmogul1]

I've not read NegPy code, but ℓ=3 band is restoring pixel values from the same pixel. This could restore real grain, but if that pixel is fully covered by dust (which happens more often than not), then ICE recovers the value only from the surrounding pixels, leading to no grain. I think that is what is happening in the attached image above. I would say dither is critical part of the Digital ICE program.

It looks like the port done by Claude removed an important piece of the whole openice code.