Achieving Digital Color Accuracy in Fashion | PI Apparel

Good afternoon and welcome in this panel discussion. We will be talking about achieving digital color accuracy, and I will be speaking with experts in this field. I am Silvia Šemčuk. I am a CEO of Fashionista. What we are doing is AI generative tool to create DXF patterns, and I will also let introduce my friends here. So let's start with Michela.

Hello everyone, I'm very happy to be here. My name is Michela Jauch and I am 3D team lead at Odlo.

This is Salvatore Tucci here, product manager in Datacolor.

Yes, and my name is Gerd, I'm the co-founder of DMX, and basically color is everything in our life.

I think we have another mic here also. Okay, cool. So I wanted to start with the first question: what exactly do we mean by digital color accuracy, and how does this define a shift across design, development, and product?

Well, color accuracy is really to find out how how much is the fidelity is for the color to be reproduced against a standard, against a reference. So how much is the quality achieved when we try to reproduce? And I have to say, this is such an important topic because Datacolor itself is basing all the business related to color accuracy. And when we talk about color accuracy, we are not talking about just a way to represent the color. It's actually a way to also formulate a color, and a way to measure the color. We do basically start from the measurement with professional instruments, and that basically use proper conditions to measure, because there are different factors that affect color accuracy. You can measure the color in wrong conditions from using a wrong device, not calibrated, for example. So there are many, many factors that affect color accuracy.

And how brands can ensure that the color is accurate? How can they identify that? How they can measure it? And how can they make sure that between the teams, yes, it is working correctly?

Well, it's a good point. The point is, first of all, the color needs to be digitalized. Once it's digitalized and stored in a centralized place, basically where everybody can access to the same reference. That sounds silly, but it happens. It happens basically that the suppliers is validating the colors against the wrong reference. It happens. And basically, um, the way to do it is to ensure it is to put in a proper software that is, as I said, make it available to all the supply chains. So we not just measure, we just keep it safe and consistent amongst brands, retailers, and suppliers. When I say suppliers, another important topic is that we have to include all the supply chains, meaning not just the mills, the dye house, but also actually the manufacturer. Manufacturer, in my opinion, are very key part of this game because they are the one that has to check the quality of the incoming goods, for example. So everybody has to access basically the same couple fabric and colors to be sure that they are evaluated evaluating the color. How do they evaluate the colors? Well, basically, I don't want to be so technical, but I mean, it's a matter of evaluating the Delta E, for example, in different color spaces. As far as the Delta is lower, as much the currency is achieved. And well, I can argue.

Yeah, I know, I know that it's your topic. And Gerd, What is your take on this? And also, Michela, from the brand perspective later.

Yes, I think you're absolutely right with what you're saying. I think the important bit really is to make colours in the right way accessible at the right touchpoint, right? And this can be, of course, if we go on the physical side, it's of course, it can be the dye house, it can be the printer. But also, it can be the retouching people at the end of the chain, right? Because this is where also a lot of problems happen. And then, of course, it's also— we need to have the same color also on the digital pass. So if we go into DPC, we also need the colors. Because if we don't get this right, then in the end, our rendering looks wrong compared to the physical product. Then we have 3 different things. We have a physical product. We have a retouched version photo. And we have a rendering. And this is what we need to streamline, and that's very important also really to gain momentum with the digital things, right? Because otherwise we will never be taken seriously in the end, right? Because it just doesn't look right.

I agree with you, Michaela.

Okay, so I will give practical use cases. I totally agree with both of you gentlemen. I had a case that basically design team handed us over approved color palettes, and then I rendered all the garments with our preset, with our custom lights. And when I compared the final render with the Pantone swatch, it looked totally different. So we basically got to the point that we cannot take blindly the color palette from the design team, even though it has been approved, verified, everything. We have to render every single color and compare the render. So that's one practical point.

And I think it's also very important, and this is why we basically do what we do at DMX, is you have to have a single source of truth for the color. You have to have one system where you can always go back and check and know how you can see the color, you get the right values, they're always mathematically converted the right way. And then basically you have a chance to go there, right? Otherwise, you never know who's right, right? And that's the important bit, really.

Exactly. And then one more thing, what you also mentioned before, is the finalization. So basically that the render, because we are going with all three directly B2C facing. So we have their mix of 3D, photo shoot, sometimes videos, and they all have different lighting. They all have different color shade. So basically, then it goes to the retoucher, post-production. He looks on all those assets, and he's having the sample, the physical sample in hand, and comparing and adjusting all of them in post-production.

And I think here, really, it's very important. So the problem with the physical sample and the retoucher, also, of course, with designers looking at colors, is, When we as humans look at colors, it's always a subjective impression, right? We never know really in terms of numbers what we are seeing. And we cannot express that. And even if you have a retoucher and he is working 10 hours on different garments, he might be quite precise the first hour in his work. For the rest, he will drop. And he will get less and less accurate, even if he's very, very talented, because he just gets tired. The solving of this can only be done if we not only give them a physical product, but also give them something digital that reflects the appearance of the material and the color. Because sometimes even the materials are so complex that the color is not alone sufficient. You know, if you are talking about a leather or something, you can stroke it like this, or you can stroke it like this, and you have completely different colors, right? And still, you want your customers to have a consistent color journey across their e-commerce, for example. And this is why sometimes, you know, it's also about the digital twins, getting these right and getting the appearance of those right for the whole chain to act on. And now comes AI, right? So that's basically on top.

You know, after what you said, that one was meant to be the last question, but it's It's the one that is coming to my mind immediately. So given that you said someone who is working on that, someone who is adjusting these images, of course, after 10 hours of work can be that accurate. Is AI the solution here? Can AI tools help with that and automate this process, make it more precise?

Well, I haven't seen something doing it correctly by now. But I do strongly believe that this is the way forward because actually, I mean, we are talking about comparing two things and then exchanging colors in a natural way. And, you know, by what we see now, what's possible with AI, this should be a task that can be solved. I think the challenge is, again, you need the right data, right? So you need, first of all, a bottom line where you know what you see and what you have in your data is correct before you can train an AI to do the job in a correct way. And it's an interesting field. So I believe we will get there, right? And maybe not even AI is really needed. It's also really about other proper tools of doing it for somebody who's doing retouching or somebody in 3D doing it. But yeah, it's an interesting approach.

So talking about one source of truth, because this is probably the most important thing and this is something that have to be set up. So what processes or checkpoints in the process are essential to make sure that the process is consistent and maybe a kind of quality standards when I was working in Timberland, we had virtual quality standards that were also talking about color accuracy. We were using, of course, Pantone, like I assume every fashion company is doing. And we had our standards in lighting, in colors, in how that should look like in 3D. So what is your take on that?

Well, I think that basically, yes, you can start from Pantone, but then your standard may become the one that your supplier is able to achieve. So you have to start from what is feasible, what is really come from the lab or from the actual dye, and that it become basically the new standard, the new version, the digital version. That's going to be— the problem is at that point to promote it as a new standard. But indeed, as we also discussed before, basically to be sure that the software allows you to centralize this information to make it available to all the supply chain. The, the reason why I mentioned that is because, I mean, we have been talking about AI, we have been talking about a way different applications for AI and We are using that also in the algorithm. We already use a state-of-the-art algorithm in order to create and to simulate a color. And this is a starting point. So we have to try to simulate the Pantone. That's fine. But then the way to do it needs to be improved. And the AI can help a lot. I'm talking about the formulation of the color. I'm talking about a way to represent that color from the supplier side. And once it's— I mean, there is a lot of improvement there. And once it's very close to what you get, what you want to achieve on the brand side, the other point is to make it available in a collaboration platform. That is—

And given that, you know, you may have the standards, you may have your Pantone, but then every fabric at the end is reacting differently when it's dyed. So is there any way to, to have it consistent? To understand, like, are there already tools in the market that will let you to make sure that the color will look exactly this way?

Well, of course you can correct, you can correct, and you can try to learn. I mean, there's machine learning out there that allows to learn from the correction that you keep going. And we are experimenting this. We are experimenting how many times we correct that color to resemble the originating standard. Yes, the answer is yes. You actually have some, something like that. And this is actually, uh, on the field and is constantly improving, leveraging the AI as well.

And this one is to, to Michela, because I wanted to, to hear it from someone from the fashion company. So how, when you started your digital journey in 3D, when the topic of digital color accuracy came out, because I assume it wasn't the first problem you, you had to face. So at what point of the journey this one started to be the roadblocker?

So it usually starts with problems. So all of a sudden you're having returns, clients are complaining that they saw on the website a certain color and what they receive looks totally different, claims. And then of course people start to trace where the problem is. So we actually even found out that the first problem was at the design team and they were using different color profile. Because in the old times, the approvals were happening based on printed designs. So they were printing everything out and they had printing color profile in the Illustrator. So they already from the start, it was wrong. And then of course we were tracing through the whole process and then we realized that everybody had a different color profile in their software. So this is very important. Everybody has to be in the same gamut and in the same color profile. So then when we started to really go to the root, aligned on the color profiles, we align on the gamut, and then of course we had to do some investments. So we invested in really high-end monitors. So I'm very happy to say that I'm working with a 1 billion, with a 10-bit monitor with 1 billion colors, which again, sometimes people are coming to me about, "Mikhaela, you are having this fancy monitor. I'm only having 8-bit monitors. I don't see what you see." But still, we keep on working on that one. We check occasionally on the other one as well, just to be sure that everybody sees the same thing. And then of course it has to go, it has to cascade down to the last stage when you have all the retouchers, the marketing people, everybody is on the same page, everybody's on the same profile gamut and calibrated screen.

And how long it took to, you know, to get back on the right track when you realized that that was the wrong one?

I would say roughly 1 year. And that's pretty fast because sometimes it takes a few seasons. And okay, let's say that we were getting the complaints for already 2 seasons before we started that process. So altogether from the complaints till the resolution of the problem, 2 years. But we were really on it and tracking it and exploring what devices to buy, which way to go altogether, probably. 2 years with starting the process.

That's not that long, given how it looks like in big fashion company Gert. And how does it look like from the perspective of the vendor who is providing this kind of solution? When you are working with fashion brand, how long is the process of onboarding, of achieving the consistency?

Well, it's— so there are two points here. So if we are just talking about color measurements, so using a Datacolor instrument or some other instrument to measure the color, basically it's quite quick because the measurements can be transferred into all sorts of color values, right? So we can go into different profiles, we can go into different lightings, and all that kind of stuff. I think the deeper problem or the more complex problem is, of course, if you go into material and material appearance, because here, obviously, the devices on the market are often not as precise as we need them to be. So then again, you end up with a scan of a material that doesn't look correct color-wise, and then again, you need to have a system where you can make sure that even if you start tweaking the colors, that you are tweaking in the right direction. And here also, I mean, there's another trap here a lot of people dive into. So when we talk about our different lights in fashion, we always talk about D65. So I'm quite sure if I look around here and ask you, which is your main lighting? It's D65, which is correct in fashion and in physical way of thinking. If you are going to a computer, if you go to color management on a computer, it's not D65 because everything what we use on Adobe tools and on other tools is coming from the printing industry, right? This all started on printing, and printing uses D50 as their standard lighting. So basically here is another shift that is happening. Most people don't know, so they're comparing something in completely the wrong light. And this is something where you get the suppliers, you need to get them on the right track, you need to teach them, you know, how to do it properly, and then basically you have a good chance and a quick chance to get to something that is right. And of course that's true for scanning, and that's also very true for material simulation, right? So of course we can do material simulation from weaving simulations or knitting simulations. And again, we are in a project now where we are doing yarns and weaving simulations, and we start from the actual fiber together with a partner. And again, it's very crucial you measure the fiber correctly, you gain the color, and then suddenly you can see how this works, or even on a simulated material perfectly.

So the physical simulation of material is also affecting how you see the colors.

Obviously, yes, of course. I mean, it's— and then, of course, you have very simple materials, and you have very complex materials. Sometimes you have materials that are depending on how you look at them. So you turn them, and they look different.

Yeah, I can imagine velvet, for example, is pretty hard.

Exactly. So some of these materials are pretty hard to capture. But, you know, if you have bottom line a system that keeps you on track, you can go very far.

Yeah, that's right. Salvatore, in your experience?

Well, onboarding is always quite tough. It's a matter of providing the real value. And in our case, it's quite straightforward in the sense that The moment that we provide— I mean, we have been already discussing that. My point of view, the crucial part is to make it objective and not subjective. So the moment that we have to get out of this field where everybody has his opinion, because otherwise you'll never get an agreement. And the moment that you demonstrate the value of relying on data as a moderator, as a referee, It is basically quite easy now to onboard the customer. And on top of that, basically the way we are approaching the customer is to educating the audience, educating the customer with training, certification, and real training on colors, on color visualization, color aspect, this kind of illuminance that even Gerda was talking about. So it's a matter of educating first of all. So we start with the education first of all. They have to know what we are talking about, and then we talk about values and so on.

That's a, that's a great approach, because if people understand what they have to do, it's, it's more likely they, they can succeed. Um, the last question I have is, what are the biggest systemic or technological challenges currently preventing the adoption of digital color accuracy at scale?

Well, I mean, of course, there are always challenges, right? And of course, you know, there's not an easy answer because there's not one material you are talking about, right? So as soon as the material gets complexer, it gets more complex to catch it in a digital appearance. But I would say there are not many hurdles that you cannot solve with the technology we already have, right? It's a matter of doing it. It's a matter of also accepting certain rules that just simply apply by the technology if you like it or not. That's also often—

I think this is the biggest challenge, accepting the rules.

Sometimes it is and sometimes it's not, right? So sometimes people are just, you know, They see, they believe, and then they accept. But if you don't accept it, then basically you are working against the windmill, right? Because you will not change color management in a computer, right? It will always be the way it is, right? And this is sometimes the biggest hurdle. So change is a hurdle, yes.

And let me add also the fact that it's going to be much more difficult for the mills to achieve that standard, because indeed, you can always argue argue, well, I did this right, so you gave it, no doubt. So you start basically— I mean, it's like being able to sell your product, okay? The moment that you are able to sell and to defend your— basically your result and your job, the better you are. But it's not correct, okay? It's something that indeed the brand retailers should be more strict and to require basically to the supply chain to follow the rules and to guarantee to guarantee basically quality and consistency.

And Michela, from brand point of view, the biggest challenges?

I would just add to what the gentleman has said. From the perspective of returns, what really helps is data. If it's data-driven, if you have like the percentage split, okay, we have 70% of returns due to fit, 20% due to color, 10% is other, then you have real live data.

Do you really know? Do you track it? Do you know which is color-related?

We do.

And once we had that data, we really had the force behind.

I would love to hear some more about that data because obviously it's very interesting.

Not everyone has it, right? Because we are asking our clients when they are returning the goods, what is the reason? And this was the game changer. Once we had the data, we could really go forward with some solution.

Can you give us a number? Is it really— was it really a big topic?

I just told you the number. 70% is fit, 20% is color, and 10% is other, roughly.

How much? I haven't heard.

20% color.

20% is a lot. You see, like, we are trying to solve the size topic, which is really important. And it's also one of the biggest reasons why people are returning garments. But if it's 20% because of color accuracy, It's a huge number, I guess. Exactly, exactly. Probably nobody even here expected that it's such a big number.

And I mean, obviously, it's always also complicated because we cannot control where our customer is looking at the digital assets that we are providing, right? So you don't have a lot of control. I mean, we are now looking into providing different spaces for devices that we can identify. If we know it's an iPhone, we can basically do another color process than maybe for something else. But I think what's very important, and then this is coming back to the single source of truth for everything, is that the products and colors next to each other and the difference between the colors needs to be correct. Because otherwise, you end up buying something that looks very nice in the shop online, but you unpack it and suddenly it doesn't have the same impression anymore, just because maybe one was retouched by this guy and the other was retouched by the other guy. And boom, here's your return. And that's basically, you know, a way you have to look at this.

No, but I agree, because if I have seen beautiful yellow t-shirt in e-commerce and I receive something which is completely different color, it's not anymore yellow, maybe it's brownish or it's more orange, I don't want it anymore.

And if you go to the physical side of things, so if you go to dye houses, I mean, if you have big brands, sometimes you have the same article that is dyed by 3 different dye houses. And if you don't have a perfectly concise standard for all 3 the same, right, you will never get anywhere, right? Because what we see also is a lot of practice where people come from design with an idea, then they do a lab dipping, and then the lab dipping becomes the standard. And this works only if it's, you know, one dye house. As soon as you have two, basically you have no standard anymore because it's going into maybe different directions.

And how we can keep the consistency here in manufacturing process, not only in digital?

Yeah, it's basically defining the standards upfront, also defining the rules of quality that you want to apply. Apply. It needs to be the same for everybody. And of course, I mean, we are in textiles, so, you know, we cannot clone, right? We will always have deviations. And what is very important is that we look at the direction of the deviation. So if somebody is deviating in this direction and the other one is deviating in this direction, you know, the overall thing is quite big, even if maybe in numbers they are on the same tolerance. So what we need to make sure is that if we do something in blue, that everybody stays in the same direction. And then basically, even if they deviate a little, it's still pleasant for the eye and we will accept it.

I believe also very interesting part was the one when customer see the garment on their own device, which makes a difference. So there are already ways to understand which device is used using customers, and you said you are able with your technology slightly change the way colors are presented based on device.

It's a matter, of course, of profiling, right? So there is technology to do it. Normally, so I'm not sure, we were just examining this now, is how many shop systems are really able to then switch, right? This is something new, and maybe it's a matter of just getting the picture that is displayed from another system So they need to be much more flexible in how data is displayed. But it's possible, right? It's— we are doing some tests now with a company that's a customer. And, you know, there is no total solution because, I mean, there are so many displays in the world. But on the other hand, again, you will find, and maybe you know this better than I do, that you can identify maybe 50% of your customers are using an iPhone. So why don't we, you know, give them the perfect experience on their device, right? That's something also to look at.

Yeah, makes sense. Salvatore, what is your take on this part of digital accuracy for the final customers, how they can perceive it using different devices?

A good point. I was just thinking about the answer because it's so challenging. I thought that the colleague here would have already the answer. It's so challenging that I don't think that we have an answer. I mean, indeed, the expectation— we are talking about user expectations, and they expect a good that is going to be released in a different way, from different colors and different shape, and That's, of course, is not acceptable. I don't have a concrete answer to that. The only thing that I can tell you is that indeed you prevent this by learning by mistake. So the only thing that you can do is just to understand what was wrong basically. And if you have a profile that helps you basically to learn from the mistakes, learn from the return, and then you can improve. You can understand basically whether this purchase has been done with that device or that kind, type of device and try to correct for the next time. You cannot prevent it basically. But you can learn. There is always, I mean, every time that you do a mistake, there is always a lesson to learn.

So the last question to Michela and then maybe we see if we have any questions from the audience. So one Since you started your journey with digital color accuracy, do you have some numbers to share how it dropped, like the rate of returns based on colors?

Okay, so there are still some returns. So I would say we are currently around 5%. So it dropped quite significantly. The returns which are there Right now are more due to material. So there are materials, specific touch, specific like brushed stuff and this kind of things. So of course, they look a little bit different. Also, when you are doing a photo shoot, they look slightly differently depending on the light and some other things coming into the photo shoot. So we still have some, but it dropped really a lot. So 15% drop, I have to say.