Rawtherapee Processing Challenge feedback

Rawtherapee Processing Challenge feedback

Introduction - Who this document is for ?

This Feedback Report (Retour d'Expériences : « Rex » in french) was produced for the Rawtherapee Processing Challenge, which took place in March and April 2024. The aim of this challenge was twofold: a) to allow you to familiarise yourself with Cam16 and the tools that have been integrated into the Local Adjustments "Color appearance (CAM16 & JzCzHz)" module. b) to validate the changes before merging into "dev". This feedback, « Rex », can help you to better understand Rawtherapee, its architecture and its tools, and to improve the quality of your images as quickly as possible by benefiting from the experience of others.

Who this document is for ? The target audience is anyone with an interest in image processing (I'm excluding the absolute beginner, as he or she doesn't have the minimum knowledge required to get started): beginners or experienced users, experts or not, amateur or professional photographers, developers, algorithm designers, and so on.

The images

We have chosen 6 images that are already well known on the forums. They are all 'difficult' in that simple processing is rarely satisfactory. Brief characteristics of these images, which will have a major impact on processing:

• IMG_0080.CR2: sunset, underexposed image, digital noise, high dynamic range 12 Ev.
• 2010_MONTR_033.NEF: "The blue horse". Shaded and sunlit areas, deep whites and blacks, isolated green tree, sky reduced to a small part of the image. In the centre the horse, its blue reflection and the carriage.
• S7_00463.ARW: Very under-exposed background framing the dance scene, artificial lighting, digital noise, high dynamic range 16Ev.
• B.dng: isolated house in the countryside, overall dull appearance, lack of contrast (dynamic) and color.
• IMGP2426.DNG: sunset, strong under-exposure to 'keep the sun', high Dynamic 14Ev.
• 2024-03-22_14-28-41.75_DSC5286.NEF - Quantum of the seas - blur - low contrast.

I have added 2 images outside the "Challenge" because they were cited for Cam16's use :

• hagalund-test01.DNG - Smooth highlights example,
• Sweep-sRGB_Linear_Half_Zip.tif , for Dynamic Range of 25Ev.

All these images are CC BY-NC-SA 4.0

How do you approach this problem? The how

The 6 images were processed by people with different experience. I'll leave it to you to discover - there's no "right solution" - how each of the testers treated the images. You'll be able to get ‘how and why they did ?’, and I hope to benefit from their experience to help us improve.

I've attached the 6 images. For each:

• the Raw file ;
• the thumbnail of the image as it appears unprocessed - on the left ;
• a thumbnail preview of what can be achieved - right;
• all the pp3 files - on the left, with a very brief summary of 'how' the tester did each one - on the right.

In a few cases, you'll see a 'pp3' ending in 'jd'. As part of the 'why', I have added one or two additional 'Local adjustments' to an existing treatment (arbitrary choice) for educational purposes.

Image n°1 IMG_0080.CR2

• Raw file(Common Attribution-share Alike 4.0): [1]

Original image without processing

Initial image

Image with possible processing

Image possible

Some treatments carried out by the testers

Tester A

• IMG_0080.CR2-Priort.pp3 [2]
• Treatment Summary: Exposure, Tone-Mapping, Impulse & Noise Reduction, Haze Removal, White Balance 7733K, Vibrance, Retinex, Color Appearance & Lighting, LA 1 spot Dynamic Range & Exposure, Graduated Filter.

Tester B

• IMG_0080.CR2-Nosle.pp3 [https://dr

ive.google.com/file/d/15wnkXYQCsZj7QZG6D69o_DsnYDpR-iAd/view?usp=sharing]

• Treatment Summary: Exposure, Temperature correlation, LA : a) spot 1 – Cam16 & Denoise – b) spot 2 - Denoise

Tester C

• IMG_0080.CR2-Hiram.pp3 [3]
• Treatment Summary: Noise Reduction – LA : 9 spots – Cam16 - Mosaic

Tester D

• IMG_0080.CR2-Paul.pp3 [4]
• Treatment Summary: LA : 1 Spot Cam16 - Resize

Tester E

• IMG_0080.CR2-PD1.pp3 [5]
• Treatment Summary : Exposure, Color Appearance & Lighting

Tester F

• IMG_0080.CR2-Asil.pp3 [6]
• Treatment Summary : Exposure, Tone-mapping, Noise Reduction, LA : 1 spot – Cam16 & Contrast by Detail Levels

Tester F + jd

• IMG_0080.CR2-Asil-Exclud-jd [7]
• Treatment Summary : As above (Asil), then : LA : 1 Excluding spot

Image n°2 – 2010_MONTR_033.NEF

• Raw file(Common Attribution-share Alike 4.0): [8]

Original image without processing

Initial image

Image with possible processing

Image possible

Some treatments carried out by the testers

Tester A

• 2010_MONTR_033.NEF-Asi.pp3[9]
• Treatment Summary: LA : 1 spot - Cam16

Tester B

• 2010_MONTR_033.NEF-PD1.pp3 [https://drive.google.com/file/d/1Au6D3z01y0eISLKkAIOAftv9wg7sVlud/view?usp=drive_link)
• Treatment Summary:Exposure, Tone Equalizer, Noise Reduction, Color Appearance & Lighting,

Tester C

• 2010_MONTR_033.NEF-Priort.pp3 [10]
• Treatment Summary:Exposure, Tone-mapping, Dynamic Range, Sharpening, Noise Reduction, Haze Removal, Color Toning, LA : 1 spot Cam16

Tester D

• 2010_MONTR_033.NEF-Nosle.pp3 [11]
• Treatment Summary:Exposure, Temperature correlation, LA 1 spot : Cam16 and Blur/grain & Denoise

Tester E

• 2010_MONTR_033.NEF-Hiram.pp3 [12]
• Treatment Summary:Exposure, LA – 28 spots Cam16 – 1 Blur/grain & Denoise

Tester B + jd

2010_MONTR_033.NEF-PD1-LAjd.pp3 [13]

• Treatment Summary: As PD1 then 2 Spots LA : Exposure, Tone Equalizer, Noise Reduction, Color Appearance & Lighting, LA : 2 spots Cam16, Wavelets

Image n°3 S7_00463.ARW

• Raw file(Common Attribution-share Alike 4.0): [14]

Original image without processing

Initial image

Image with possible processing

Image possible

Some treatments carried out by the testers

Tester A

• S7_00463.ARW-PD1.pp3[15]
• Treatment Summary:Exposure, Tone Equalizer, Sharpening, Color Appearance & Lighting

Tester B

• S7_00463.ARW-Nosle.pp3 [16]
• Treatment Summary:Defringe, Temperature correlation, LA 2 spots – Cam16 and Denoise

Tester C

• S7_00463.ARW-Priort.pp3 [17]
• Treatment Summary : Shadows/Highlights, Tone Equalizer, Sharpening, Noise Reduction, Defringe, Contrast by Detail Levels, LA 1 spot Cam16, Resize

Tester D

• S7_00463.ARW-Asi.pp3 [18]
• Treatment Summary : LA 1 spot Cam16 – Crop - Resize

Tester E

• S7_00463.ARW-Hiram.pp3 [19]
• Treatment Summary : LA 5 spots Cam16 – Crop – Resize – Post Resize Sharpening

The new features tested in this challenge

• The 'Source Data Adjustments' - (SDA) module with its components: a) 'Tone Response Curve & Midtones', b) 'Highlight attenuation & Levels'. In 'Advanced' mode you have access to: c) the ability to act on primaries and illuminants, d) Log encoding. Together, they form a tool similar to Abstract Profile, located at the end of the pipeline. This SDA module, located 'before' Cam16, improves the processing of images with high dynamic range or blocked shadows.
• The 'Ciecam - Cam16' part, which is a simplified version of 'Color Appearance & Lighting', is a CAM (Color Appearance Model) that takes into account not only the Cartesian aspects of photography, but also the physiological aspect (eye-brain relationship).
• You can also try out the JzCzHz experimental module in 'Advanced' mode, which is an attempt at an HDR approach.
• And of course, this challenge was (is) for some of you an opportunity to get to know Rawtherapee better.

Color appearance Cam16 and HDR

Why a new module ?: there's Sigmoid and Filmic

Communication is one of the foundations of success. In this respect, other software (paid or free) does considerably better than Rawtherapee. They have succeeded in capturing the attention of users, through tutorials, videos, forums, etc., where the response from contributors (developers, testers, etc.) is rapid.

The example of one of Rawtherapee's “competitors” is particularly striking. Based on a totally false assertion (Lab is bad), the whole thing was rewritten. This doesn't mean that the result of this system isn't good, but it sows (bad) doubt in the minds of those who haven't switched. Admittedly, Lab isn't perfect, especially when it comes to color drift when changing saturation, but RT solves it perfectly with Munsell corrections.

Reservations about Lab and its limitation to 7 or 8 Ev are unfounded. Image processing - after RGB>Lab and then Lab>RGB conversion - shows that Lab easily passes 25 Ev, more than enough to process all images. An evaluation of the dynamic-range capabilities of tools in Selective Editing

I'm not saying that Filmic (cf Log encoding) or Sigmoid are bad modules, these are tone-mappers - you'll find them in RT - , but either they use non-linear conversions that disrupt colorimetry, or they can't both release deep shadows and soften highlights.

I set out to find a credible alternative that would be as no color-destructive as possible, while at the same time being able to play on the lowlights in a linear way:

• on lowlights in a linear way
• on highlights in a parabolic way (gamma) as our eye does

This first step, based on the TRC (Tone Response Curve) and the virtual profile, has often been misunderstood (lack of communication), and yet...easily solves the problem of many images.

Over the years, I've added several additions to it:

• complete the action with a midtones slider.
• soften highlights (highlight attenuation) when they are too centered on the limit value, thus losing essential detail.
• improve dynamics processing (DR) without the need for complex or destructive functions (Log, Fourier transform, Laplacians, etc.).
• provide access (in advanced mode) to editing of primaries,illuminants and dominant colors, using a module that may not be very intuitive, but is nonetheless powerful.

Together with “Scene conditions” and "Color Appearance Cam16", this is part of an overall process that takes maximum account of the linearity of the data being processed (and to be processed), and of physiological aspects. We find the name "Scene referred and Display referred" which here are combined.

In summary, this module does not use logarithmic conversion, sigmoid, Laplacian, or Fourier transform. This does not mean that it is simple, it is different, and restores 23 to 24Ev. What RawTerapee lacks is, above all, good communication. RT seems to some to be software for engineers, while others seem to be software for photographers. This image linked to the presence of functionalities which do not exist elsewhere (or in an embryonic state) is false - you are not obliged to use wavelets, ciecam, etc., but maintained by (bad) communication internal and external.

Rawtherapee processing and interface

Each software package (Lightroom, Darktable, DxO, Rawtherapee, etc.) has a different ergonomic approach. This approach is linked to its history, to the first developers who worked on it and who have different experiences and cultures, resulting in very different systems. There is no one right approach. For the user trying to switch from one to the other, the result is confusion, even rejection, in the face of apparent complexity. Rawtherapee was created in 2006 by a single person - Gabor Horvatz - and the architecture of the system has remained the same.

New modules have been added over the last 18 years. By design, these new modules are not highlighted when the system is opened. As Rawtherapee has developed, new tools and algorithms have been added, in several existing "Tabs", or even in new Tabs, which are "behind" them. Let's take an example: a user wants to lighten the shadows in an image. Depending on their experience, they will first go to the first 'Tab' and the first 'Exposure' tools... You really need to be curious and well-informed to process the image with the Surround function in 'Color Appearance & Lighting' - Tab advanced.

The testers who took part in this 'challenge' each had their own profile (experience, culture...) - which evolved over the course of the challenge, so you learn from yourself and from others. This led to some very different image processing. There is no "one" good solution, but a process that matches each person's profile.

The impact of RT's internal design - the pipeline

Knowing at least the sequence of the main tools enables you to understand the impact of each tool on the other, and thus to induce or not induce a change in the way you approach and process an image. This summary is not intended to replace Rawpedia or the help available on the forum, or the videos, etc., nor to impose a particular method.

In order - summary of the main methods and tools :

• Demosaicing method, more or less precise, more or less adapted to noisy images ;
• Input profile: manufacturer's matrix, ICC profile or DCP;
• White balance: camera - auto – manual;
• Spot Removal ;
• Highlight recovery ;
• Noise reduction ;
• Dehaze ;
• Dynamic Range Compression ;
• Local Adjustments: which can be used in local mode or for the entire image. These include tools similar to those in the main process, as well as 'new features' (Log encoding, Cam16, JzCzHz, Original Retinex, etc.) and masks;
• Auto-match tone curve ;
• Tone response curve and the entire RGB process : Exposure, Shadows, Highlight, Black, Color toning, HSV equalizer, etc.;
• Lab process: Shadows/highlight, Lab adjustments, Graduated filter, Tone-mapping, etc. ;
• Specific tools for a more complex approach: Wavelets, Abstract profile, Color Appearance & Lighting ;
• Crop, Resize ;
• Final output conversion.

An upstream tool will not be influenced by a downstream tool, but the reverse is true. You will notice that there is no relationship between the order of operations in the pipeline and the position in the graphical interface (GUI).

A possible answer to why? Testers' comments

Tester 1

One way we can gain that kind of old-hand experience is learning how photos are developed by hand. When you view the image under the exposer, you separate out where the lighting zones of the photo exist, and determining how you are going to affect the exposure to various areas of the photo using your hands, other small hand tools used for modifying the light hitting the paper, and various timing methods. These areas are very much like Local Adjustment zones, so looking at a photo in zones from the outset definitely invites a deeper understanding of how a Local Adjustment spot could help the area in question, and affect its overall balance within the rest of the photo. These are all editorial decisions, a bit more than just knowing what every knob in the software does, because you are intending to affect how the image is perceived by acting as a director for the viewers eyes. The editorial content starts with the photographic composition, and follows the image thru exposure developement. The photo editor can tell a story with an image using all this know how, and the better they know their skills, tools, and appreciate the realities of retinal color sensation, visual cortex perception, cultural expectation, they will attain the desired result with greater ease and efficiency.

Tester 2

Frankly, I don’t see myself competent to comment as I am relatively new to RT. However, I can say that my focus is on how the picture develops from a visual perspective. So typically I start with the exposure and move on to the others, which make meaning to me. I found the module “Colour Appearance and Lighting” very useful as it was able to handle a wide range of adjustments in a single module. This was also, I understand the focus of the challenge, and I’d say I used it in almost all the pictures to get results to my satisfaction.As I am an amateur, I use the basic configuration of a laptop for my photo development, so many of the finer details I miss, as I don’t have a high-resolution monitor of a respectable size, required by someone who is a serious post-processing pro. Because of this, I don’t see how I can use some of the other modules, as I am not sure how it can help. I am, what can be called an enthusiastic amateur, and someone, who is not very knowledgeable in the technical aspects.I hope this helps. I was hesitating giving feedback as I did not see how my input could help, being at a basic level

Tester 3

As far as the tool itself is concerned, it is capable of giving results which are as good as, if not better than other software (Foss and non-Foss). The downside is that it takes more time “fiddling” compared to some of the alternatives. This is not a problem of complexity per se but complexity in so far as it affects the efficient use of the software. This goes beyond the tool itself, which could be described as CAM (Color Appearance Model) with a bit of Log encoding. However, we also have Log Encoding with a bit of CAM in local adjustments and CAM in the main menu, each with a set of sliders that can react differently depending on the tool.

Tester 4

I only test if there are bugs.

Tester 5

I gathered from the discussion and the documentation in Rawpedia on the Color Appearance & Lighting (CAL) that it is a single module that can be used to process most of the pictures. No other module is required The exceptions are with some specialized modules like Crop or Spot removal etc. Is it true, yes I found it so when I used it once again on three images (with different conditions) and purposefully did not touch any other. The output IS satisfying for all of them.For people like me, who process an image visually, and not technically, I found the Rawpedia article wanting, it wanders far too deep into the technicalities. It does not talk about the various sliders (the why) and what they do and maybe gives situations when they could/should be used. I was left experimenting, not knowing what to expect. Except in the beginning, when some examples were given, the rest of the article was all Greek and Latin for me. (excuse the expression). It is from these examples I got the idea of the ‘power’ of the module.The Rawpedia article needs to change considerably if you want the user to use the module effectively, and keep the technicalities in addendum as links. Overall it is an excellent module, and when used effectively, would obviate the need to use any other module (except for the exceptions stated earlier), as long as the user wants to process a picture with a balanced tonal quality. For artistic expression, other modules would be required. I used the latest version in zip format, which is downloaded from the dev page. Hope my rambling is of use to you and other members

Tester 6

I interpreted the challenge to be mostly about local edits and the ciecam module. I use these for my own work so it wasn’t much of a change although I refrained from using other tools , such as tone equalizer, in one or two occasions. When I have an image to process i look at it and quickly assess what obvious issues it may have. Personally it’s often exposure and dynamic range management that needs adjustment. This comes largely from the digital requirement to expose to avoid clipping. Several of the images in this challenge had this issue to manage. I rarely strive to change the image in terms of relative colour relationships and focus on exposure and detail. To get a feel for the technical limitations of the file I often dramatically change exposure to observe clipping, noise and other issues. I also often do this using other tools such as Color Appearance. This gives me some understanding how the particular image reacts to particular tools and hints and when I need to stack multiple tools that handle the same issue but behave slightly differently. My starting point is often auto match tone curve. I’ve set my own cameras to a modest sooc profile so it gives good results with auto matched. I usually leave this on as a sort of look curve. I find it gives good results for overall tone and even leave it on when editing files from other people. Naturally I change the curve when needed but only after finding that it interacts badly with other tools. For the typical underexposed image I know that setting « Color Appearance > Scene Conditions > Surround to dim » does a good job as an initial boost of shadows. Dark rarely looks good but I’ve found photos where it does so I often check. For the horse image I needed to boost shadows quite a bit both changing surround and using source data adjustment, this always affects contrast in a problematic way. To counter this there is a variety of options but in this instance i choose Ciecam contrast. As the title of the file suggests and the nature of shaded areas on a sunny day determine the horse was now a bit too blue. This could be solved by chromatic adjustments of the shaded areas. I found that making the horse too warm looked odd so maintained a bluish tint. When files have been pushed this much noise tend to become an issue which I tackled via a local edit noise reduction tool. The other images followed a similar approach with changes in how noise was handled and how much dynamic range was flattened. The goal with all the edits was to make the images look good without radically changing the look or atmosphere. In the dancers image I left the people almost black and focused on good tones and contrast on the dancers themselves.