Any love for GIMP ?

Started 3 months ago | Discussions thread
charlyw64 Regular Member • Posts: 450
Re: Missing the point--what GIMP cannot do

His Dudeness wrote:

charlyw64 wrote:

His Dudeness wrote:

CAcreeks wrote:

(Darktable tutorials recommend starting with exposure compensation. Is this standard?)

The scene-referred workflow is centered around nailing middle gray. filmic keeps middle gray constant. So, it makes sense to adjust exposure quite early. So is white balance, as it is a fulcrum of color grading in the color balance module.

And therein lies the whole problem of that approach - it is completely counter intuitive to settle on a color grading unless you are color grading video where each frame must match the settled look.

In photography the frames can be widely differently exposed to begin with, even when I am shooting several aspects of the same subject it's often the lighting that is intentionally different and thus having to settle on a color grading makes no sense whatsoever for me, and neither does fixing middle gray -

Sounds like you think that the scene-referred workflow is about treating a set/series of photos like a movie It is not. Of course, each photo is processed independently.

And yet you need to decide about the color grading first although it happens almost last (and highly flawed) rather late in the pipeline. That approach is needed for video, for photography it is blatantly wrong as it has no benefits and the drawbacks are massive because it puts technical correctness over visual representation! You can not apply noise reduction, you can't judge sharpening (not that the available sharpening is any good - it too is a snake oil pretending that deconvolution can be used without knowing the point spread function - that only results in terrible artefacts).

At a glance, the (technical) point of the scene-referred workflow is to keep the data in the processing pipeline as long as possible in a physically meaningful representation (linear encoding and unbounded dynamic range).

Physically correct is not visually correct. The human vision is a bound system and it doesn't help if you edit in unbound space and thus have one late transition into the visual representation because then you need to correct too much problems you created earlier which you didn't notice. A module like Filmic which needs a "let's repair what's gone wrong" setting is a misconception and makes the software that relies on this mess a useless experiment. What happens if you sharpen in edge cases (colours which can't be visually represented), what happens if you have noise in such areas when you apply noise reduction? The problem is that most edits will created areas that are not visually sane.

Many tools can do processing on linear captured data - but these all invariably lead to crappy looking HDR tone mapped results which these tools are acknowledging because they are marketed as HDR creation tools.

Algorithms in this section of the pipeline can benefit from this representation in different ways.

And yet these algorithms are working on data that doesn't account for the human vision and how it reacts. Thus there is no benefit to a human, only image analysis software would benefit in edge cases - as soon as the human vision is concerned then that approach is fatally flawed as these algorithms need to individually account for the intricacies of the human vision and display technology to be useful - they don't and thus they are prone to create technically correct but horribly looking intermediate results which need too much repair to be worthwhile,

The filmic module does the tone mapping from an unbounded dynamic range to the bounded dynamic range of the display,

Wrong approach, tone mapping is the wrong tool, it undoes a lot of the previous decisions and tries to repair the mess of meaningless data the unbound source has created. It is compressing the data in a horrible way, especially if you have - because of the messy algorithms used before that module kicks in - created over saturated false colours. Heck, even the sensor of the camera captures colors which the human vision can not distinguish. How does this inept mess of a tone map determine how to "map" these colors into a visible spectrum - which of these colors need to be distinguishable because they are the result of sharpening for example. It can't because the translation to a human vision constraint is happening far too late in the pipeline!

Edit: Yes, these problems can be severe and they can be easily seen because Darktable users do complain a lot about how hard editing to taste has become - often giving up on the image in question (instead of giving up on the software which would be the correct thing)..

mapping middle gray in scene-reffered to middle gray in display-referred.

Display referred is as incorrect and puts greater constraints on the image than are sensible. In all other editing software you are working with a visual referred system, not tied to a specific display. I use a wide gamut monitor but may want to edit with printing constraints - that is possible in any other software but in Darktable it's practically impossible because of the too late transformation which undoes a lot of the decisions I made earlier,

Hence, middle gray is basically the fulcrum of the tone mapping. This is the simple reason, why it makes sense to define middle gray in the particular photo early.

Middle gray though is not what any edit is centering about. Technically correct but from the standpoint of the user visually incorrect.

...he resorts to foul name calling [..] discourage anyone putting in much time into that crap. ..

Maybe, you should rethink your wording as well.

I am formulating an opinion, he is seemingly trying to formulate a scientific argument why his approach is so superior - so he says, this is my algorithm, it's correct for technical reason a, b and c and every other software is crap because it is crap. That's a snake oil merchant!

Post (hide subjects) Posted by
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark MMy threads
Color scheme? Blue / Yellow