Les mensonges de la presse

[trotter]

Héhé, ça y est, vous pensez petit à petit que les journaux sont peu fiables et les scientifiques suivent l'argent !

A 60 ans vous en serez convaincu mais les jeunes vous dirons "n'importe quoi papi t'es un complotiste d’extrême droite!!!".

[Nemau]

Que ~90% des médias d'information/opinion sont peu fiables on le pense déjà, depuis longtemps. ^^

[trotter]

Dans une émission de Mediapart un chiffre de "morts liés au climat" qui ne sont pas des morts liés au climat :


Et le graphique a été tronqué sur la gauche :


C'est fou de faire ça à l'heure d'internet, c'est donner des points facilement aux climatosceptiques.

"Une très forte augmentation ces 2 dernières années" qu'elle dit bordel.

[trotter]

Vidéo sur les couleurs dans les jeux vidéo :


J'ai commencé à regarder avec un petit "mouais..." au fur et à mesure qu'elle parlait car les couleurs qu'elle montre ne me convainquaient pas.

Elle se fait défoncer :

There's a lot more complexity to color than you present in this video. I think you're falling into the dunning-kruger effect. ACES (Academy Color Encoding System) is not a tonemapper. It's a color encoding system (it's in the name). There's a lot of information in this video that simply doesn't make sense and I'm concerned it's misleading a lot of people.

Tonemapping is all about taking wide dynamic range and represnting it faithfully in a low dynamic range environment. Throughout the video you seem to mistake colorspace transforms for tonemapping and vice vera.

Tonemapping is a very subjective process. Note how you prefer your custom tonemapper over Unreal's default one, that alone is proof that it's subjective. Without tone mapping, lumanance values would be clipped when being transformed from larger to smaller dynamic ranges.

A colorspace transform is used to transform colour between two colorspaces; when going from smaller to larger colorspaces, there is no 'preference' there, only objective truth. For example, you can do a colorspace transform between Rec709 and Rec2020.

But what does it mean to do a colorspace transform?

The keyword is relativity. It is ALL relative.

You have failed to highlight a cruicial fact; values only go between 0 and 1, with 0 being off and 1 being fully on. ACES, Rec2020, P3, Rec709... there's no difference between the data. It's all a range of 0 to 1. So if it's all 0 to 1, then what the fuck is going on?

It's relativity. Rec709 is a technical standard; it is simply a reference point. That's important - it's a guide, a goal, something to strive towards achieving. Rec709 defines specific color coordinates which can be measured IRL using a spectrophotometer. For example, you could take pure green (for example, RGB 0/1/0) and display it on a Rec709 display. You then measure the light emiting from the display with a spectrophotometer. If the measured result matches the Rec709 standard, congrats, you have a perfect Rec709 green.

But the thing is, what if you display RGB 0/1/0 on a Rec2020 display? Rec2020 is supposed to cover a much broader range of colors, right? Correct. If you put 0/1/0 on a perfect Rec2020 display and measure it, the reading will match the Rec2020 standard.

What gives? Suddenly all our color values have no meaning at all and it's just up to the display? Well... yes. That's right.

A colorspace transform is a way of mapping that 0/1/0 between colorspaces. Let's say we have that pure RGB 0/1/0 green in Rec709, but we want to display it on a Rec2020 display. What you do is transform the 0/1/0 into Rec2020. What ends up happening is 0/1/0 now becomes (for example, not correct) 0/0.6/0. Now the green is lower. This makes sense; Rec2020 goes much more saturated, so to match Rec709, you need to bring the green down. Now if you put the transformed green on a Rec2020 display beside a full 0/1/0 green on a Rec709 display they will match. It's ALL about relativity. And when you go from a LARGE space like Rec2020 pure green down to Rec709, you need to start making decisions about how that maps out, and that's where you start to enter tha land of tone mapping. Making a LARGE dynamic range fit into a LOW dynamic range in a way that represents the nature of how it looked in the large dynamic range.

With this new information I want to revisit ACES. There are MANY colorspaces that are part of the Academy Color Encoding System. ACES 2065-1 has 100% coverage of all wavelengths in our reality; actually it extends beyond it significantly. If you had a 0/1/0 in ACES, no display in the universe could ever display it. The point of ACES2065-1 is an intermediate container. It's objective. You slam all your different color spaces into ACES, and then transform out of it in a unified way. You never WORK in ACES, you only conform and store color in ACES. They tried to make ACES usable in a workflow with AP1, but AP1 is so close to Rec2020 that colorsists prefer Rec2020 since that's what people with Rec2020 displays will be watching.

But I just want to bring attention to one finaly interesting thing: Pointer's gamut. Pointer's gamut is an irregular gamut of colours that exist in our real world. That is to say, Dr Michael Pointer went out and measured real colors reflected from the objects around us. The result is a wavy gamut that represents the color we see most of the time. The only time you'll see colors ourside of Pointers gamut is with things like LEDs where they can produce a specific and incredibly saturated wavelength of light.

Now compare Pointer's gamut to Rec2020. Rec2020 covers colors that you don't really see in real life. In fact P3 is the closest match you'll get to "real life" colors, and P3 is the cinema standard.

And this really highlights my opinion on this entire matter: do you have materials in your scene that are producing colors outside of Pointer's gamut? Because if you do, of course you need a tonemapper to fix it. You're producing unrealistic colors! No wonder your image looks unrealistic even after tonemapping. Your use of postprocessing, color correction, luts, won't fix the root issue; your scene is producing colors in a scene referred world (aka linear) which is then tonemapped to look more reasonable.

Also fun fact, colorists spend their time grading in linear since it's how light actually works. So if you have an issue with how your image is looking, you need to start paying attention to the linear scene reffered color instead of the result down the chain of tonemapping and correcting.

There's so much more I could say about this since it's so deep and complex, and this is truly just scratching the surface. face-blue-wide-eyes