Request | Quick Batch Transform Bit Depth

[musicvid]

Hi, first time posting here, and an old film lab pro at Technicolor and Eastman.
The best-kept secret of Gigapixel, and a feature that is still in its infancy industry wide, is the ability of AI to uplift 8 bits per channel to 16 just by “Saving As.”

Bit depth uplifting, or “filling in the tweeners,” has been a dream of internet alchemists and hobbyists for decades. The most common fantasy is being able to turn 16 million colors (8bpc) into 1 Billion colors (10 bpc) through a physical transformation that somehow doesn’t follow Newton’s Laws of Physics. Being able to do so would increase dynamic range, saturate the colors, and challenge our senses as if simulating a true 10 bit display.

The biggest rookie error, and it’s pretty common, is to assume that encoding an 8 bit image into a 10+ bit format will “fill in” those fugitive colors. That assumption is mathematically wrong, In truth, it is the same as pouring 8 gallons of water into a 10 gallon bucket. I call it “adding air.”

The beauty of Gigapixel is AI’s ability to actually double the number of output bits by saving to a 16 bit format, Increasing the number of f/stops and available colors, with the bonus effect of eliminating banding.

So, in a scenario that could only be imagined before, Gigapixel can generate this result, by opening the uplifted image in Photoshop 16 bit float space, grading, and downsampling to 8 bit integer for delivery.

Feature Suggestion: Add a lean batch processing mode that would upscale and transform bit depth, and very little else, and would handle PNG Sequences of maybe 100 or so to start. Speed, of course, is the goal.

Not being sure what is going on with the other two Topaz programs along these lines, I welcome any discussion. Note that this topic is not the same as scanning image at 16+ bit float depth.
Mark

[AND-E]

Hi. 8-bit to 16-bit RAW conversion isn’t hidden, this is an automatic feature of Topaz Gigapixel and Photo AI.

Indeed Topaz, incorporated their jpeg to RAW conversion technology into both applications from the beginning

[musicvid]

Hi @AND-E , Thank you, and I think you missed my meaning. You are correct that nothing is being “concealed” from us; however, this is an important new feature for the phototechnical industry, and one that is almost completely undocumented by Topaz, save for one brief line in the user docs.

Quite frankly, from a commercial industry perspective of over 50 years, this thing needs to be exploited. I understand Adobe is just getting there, too. My colleagues are still largely unaware.

This topic then, is an attempt to stimulate a discussion of its technical aspects for future development directions and eventual adoption for video via image sequencing, when and if such a thing becomes practical from processing efficiency improvements.

In that sense, if there is any engineering documentation from Topaz you would share or point me to, I would like to read and share it, too. Unfortunately, real public knowledge on the subject seems to have been overwashed by the masses of disinformation being spouted on the internet by uneducated hobbyists, commonly known as the Dunning-Kreuger effect.

Feel free to post your examples.

[andymagee-52287]

It can do this going from jpeg to tiff as well. I’m not sure I can actually see any practical benefit from this, but it does happen. Do we know that it’s not just simply doing a linear interpolation of the missing values versus some AI perceptual rendition?

[musicvid]

The input format doesn’t matter. Everything is decompressed to raw bits for processing.
Tiff and Png output formats support 16 bit.
The practical benefit is that it is a gold mine for post production professionals.
There is no such thing as a linear interpolation of bits into a larger float space, since that function lies in the frequency, not the spatial domain. That little mathematical disconnect seems to attract thousands of internet kiddos who lay claim to such alchemy

If you have photo science experience, see my histogram full of holes, called banding, take note of the bandwidth numbers, and read my “bucket” analogy.

Maybe a Topaz engineer will jump in at some point and illuminate. Until then . . .