Hacker News Reader

Comment by giuliomagnifico | Hacker News Reader

You can’t not post-process a photo, because you can’t use/view the RAW file captured by the sensor, or better, only the company who build the RAW/DAC conversion can issue the “codec” to open the RAW file.

The issue is “how many interpolations/adjustments are these companies making when converting the RAW to an image file?” Too much according to the article.

View on HN

I think it's important to distinguish between signal processing techniques like debayering, auto-leveling etc and "we're using machine learning to identify humans in a picture and adjust their contrast". We can have the former without the latter, they're not all just "post-processing".

This is the type of pedantry which is technically correct, but contributes nothing to the discussion.

It seems like an oversimplification to imply all processing exists on a gradient with no significant lines we can draw between one type and another. E.g. content identification has a pretty clear qualitative difference from whatever "dumb" mappings do to get from RAW to values in some color space. One isn't just "more processing" and the other "less processing", colloquially.

Android Camera2 APIs allow RAW capture. But images still require a bit of magic to "de-bayerize" added color filters and noise ;)

https://en.wikipedia.org/wiki/Demosaicing

It’s not just a matter of degree or quantity of post processing. Some techniques are qualitatively different than others.

Reply on news.ycombinator.com