Quick journey on the measurement hell
Nowadays, more and more products and services are purchased via the Internet. This year retail e-commerce sales amounted to 4.2 trillion US dollars. This is more than twice bigger compared to 2016 and if this growth rate remains stable, it will be 6.5 trillion US dollars in 2023 (according to statista.com).
Online shopping has become a secure, fast, and convenient way to buy stuff and pandemic makes this option much more preferable than offline shopping. So, it is pretty sensible not only to pay for your Netflix subscription but for a new pair of shoes as well, right? Almost any shoe manufacturer in the world has its own online store where customers can buy any pair with a few clicks. It seems like a very straightforward action except for one tricky point — you have to enter a very important parameter on the store site — your shoe size.
We at Neatsy AI know a lot about the difficulties and quirks associated with choosing the right size of a shoe. We are developing a precise and user-friendly tool which can reduce all difficulties and ambiguities to help customers to choose a perfect size every time. In this article, I want to cover some problems and important aspects of foot measuring which we discovered during our research.
Let’s start with a simple question — do you know your shoe size? I suppose that if you had bought shoes before, some number would pop into your mind. Some number which you told an assistant in an offline store when they ask you about your size. For example, 42 or 9. But what does this number actually mean and can it be helpful for ordering shoes on the Internet?
What is foot length?
Well, at first glance, the concept of “shoe size number” seems quite simple — it should be some conversion from your foot length in centimeters to some numerical range with a fixed step size. But here things get complicated. Your foot is a non-trivial 3D object and there is no simple definition of “foot length”.
There are at least two common ways that are broadly used in the industry. The first one — the length of a fitted rectangle around your feet. This is quite an intuitive method to make a measurement and used by many retailers, e.g. Adidas.
The second one — the distance from the heel to the tip of the big toe. This approach is slightly different from the previous one and the final measurement can be bigger by up to 2 size units. Heel-Toe length is also used by some shoe manufacturers such as Under Armour.
In this way, making correct measurements at home is not such a simple task due to a lot of factors — the position of your foot, choosing points on your feet to make a measurement, an observational error, variability of human foot because of the position, pressure, time of the day and etc. The final error which you can get leads to an incorrect shoe size choosing.
We make a lot of effort to achieve very high measurement accuracy. There is another article here about techniques that we use to produce stable and consistent measurements and how we develop them. However, despite the fact that high precision is a very important part of the measurement process, itself alone it is not a panacea.
What is size unit?
Let’s imagine that we overcome this problem somehow and get a “foot length”, whatever it means. Are we ready to buy a new pair of shoes via an online store? Well, we can’t order clothes using centimeters as a size parameter — we should convert centimeters to the special size unit beforehand. Unfortunately, there is no one main size unit — there are plenty of numerical ranges that are treated differently by manufacturers. There are also some specific nuances in various countries.
The most common units are EU (Europe), UK (United Kingdom), US (United States), JP\CN (Japan and China). As you can see on the scheme above — there is no clear one-to-one correspondence between them even on this idealized model. If you compute, for example, the EU size of your foot, there is no guarantee that you can properly convert it to another size unit. The best thing you can do is to start from the beginning — to measure the length in centimeters and convert it using a standard formula.
The situation gets worse when you notice that there are some modifications of units. For example, there is no simple US — there are at least three versions of US — US Men, US Women, and US kids. All these units are significantly different but for whatever reason, online stores treat them as one single unit.
This user experience leads to careless mistakes and pushes a customer to make some efforts to select the right size even if he is absolutely sure about his foot length. Other size units seem to be more user-friendly but only at first glance.
The two tables above provide information about correspondences centimeters-to-UK and UK-to-EU in the most popular shoe brands — Puma, Adidas, and Nike. Let’s take a closer look at these tables and find some interesting moments.
- Despite the fact that there is a standard method to convert centimeters to UK, no retailer follows this convention and has its unique size charts. This applies not only to UK but to any other size unit.
- Different UK sizes can correspond to the same centimeters value. E.g. 24 centimeters in Nike Men can be either 5 or 5.5 UK.
- Some brands separate units for men and for women. In the ideal world, only US is a gender-specific unit and other units should be the same for men and women. But for some reason, Nike uses different EU units for different genders.
- There is a lack of consistency not only in conversion from centimeters. Conversion from one size unit to another is also fuzzy and deviation can reach up to 1 step.
- The step between adjacent units can be not only 1/2. For example, Adidas uses a minimal step equal to 1/3 .
The icing on the cake — different regions can have some local-specific variation of size unit. For example, EU transforms into DE in Germany, FR in France and RU in Russia. All these units do not have to be equal. E.g. RU size is EU size minus 1 and this can be very frustrating for customers (see select element for Russia version of puma.com below)
To summarize, the concept of unified size units seems very convenient and useful but the current implementation of this idea is an absolute mess. In order to overcome this problem, we have developed the flexible conversion subsystem which is aware of all these tricky moments and can find not an abstract size unit but an appropriate number for each specific shoe model.
What about height, width and other foot parameters?
If all feet on the planet were proportional and similar, the length of a foot would be a sufficient parameter to find a perfect size. For better or worse this is not true — each foot is unique and has its own 3D shape. Two feet with equal length can have different widths and heights, various levels of pronation or supination, foot fullness, etc (we have gathered a huge collection with all types of feet, so we know this for sure). All these properties have a significant impact on the comfortability of shoes wearing, so using only the length of a foot to choose new sneakers is an error-prone way.
Does humanity handle this situation somehow? Well, there were some attempts to create a method, which will consider more parameters of a human’s foot. For example, in 1927 Charles Brannock invented “The Brannock Device”. This is a special ruler adopted to determine three foot parameters — heel-to-toe length, arch length (heel-to-ball), and width.
The construction of this tool is pretty simple — you should just stand on the ruler, move the heel-to-ball slider, the width slider, and determine relevant values. After that, by using a special table you can find US size and a special foot width parameter. This width parameter works in the same way as length. Each width is separated by a distance of 3/16 of an inch. There are 9 widths in the US system: AAA, AA, A, B, C, D, E, EE, and EEE.
According to the official site brannock.com, “The Brannock Device dramatically improved the accuracy of a foot measurement, to 95–96 percent right”. It sounds really good, however, this solution has its own cons:
- The tool determines only two parameters — length and width. All other foot characteristics remain unmeasured. Moreover, the length is defined only as the maximum value of heel-to-toe length and arch length. This is not a very neat way to combine these properties together.
- The Brannock Device is a big full-metal physical object with a weight of up to 1 kg (~2 pounds) and with a cost of up to 100$. It is a significant obstacle for wide-spreading among customers because much more people prefer to just visit an offline store and try a real pair of shoes instead of buying this mechanism.
- Even if this device can measure width incredibly accurately, there are very few models that have a customizable width parameter. Furthermore, models with selectable width parameters more often have only two versions — normal and extra-wide — instead of 9 units from the brannock.
The last idea leads us to an understanding of a deeper problem: even if we perfectly measured the shape of your feet, we could not apply this information to choose a size on the online store, because there are not enough details about shoe sewing features and customizable options to select.
What are shoe sewing features?
Just like a human foot, a shoe has its unique inner 3D shape. Different shoes may vary in length, width, height, inner sole surface, beak shape of the shoe, etc. Unfortunately, in most cases, it is impossible to properly customize these properties on the online store and the only available option for a user is a single size unit, which often corresponds only to the length of the shoe. The consequence is that there is a nonzero probability that shoes will not fit you at all. In some cases, you can see the incompatibility of two shoe models with the naked eye.
The animation below highlights differences between widths of the two Adidas models. As you can see, the lengths of these shoes are absolutely the same (look at the highest and the lowest points of sneakers), so the sizes of these models are the same too, however, the difference in width may be up to 1 cm.
There is no common database with such information about every existing shoe model in the world. So, we can try to solve this problem by constructing a detailed 3D shape of a human foot and then order the production of a unique personal pair of shoes that will just perfectly fit. This approach can be too expensive for both manufacturers and customers. However, it can work in situations when a shoe brand has a narrow product line with high demands for fitting.
For example, Bauer sells professional hockey skates. It is very important for a skate to perfectly fit a player’s foot, otherwise, the feet get smashed quickly during the game. Therefore this company provides the foot scan service — you can visit a special offline store from Bauer’s partner and scan your feet with a big hardware scanner (the technology was developed by Volumental).
The constructed 3D scan allows you to buy the best skate for you. This method does not work just as well for an everyday shopping scenario because the scanner itself is a very expensive device and you as a customer do not want to buy a new pair of professional skates each week. However, if you are a player and take care of your equipment and health this can be the best way for you.
It should be noted that the quality of constructed 3D feet models via this scanner is very good. Our whole team visited a special offline store with such a scanner, got several scans of our own feet, and were really impressed. The magic is in the special sensors called Structured-light 3D cameras, installed in the scanner. This camera can create “3D photos” of the objects leveraging properties of infrared light.
Really cool fact — modern versions of the iPhone have exactly the same Structured-light 3D camera called “TrueDepth”. Apple uses this sensor for a user authentication method called FaceID, so it is very likely that you are already familiar with this technology. This circumstance inspired us to base our product on this iPhone's TrueDepth camera in order to create a high precision measurement of a human’s foot.
Implementing an efficient scanning algorithm is quite challenging and you can read more about hardcore implementation details in another article.
Well, using this amazing sensor we can reconstruct a perfect digital image of the foot. The challenge that remains is to gather a database of inner shapes of each shoe model.
How to create a meaningful digital representation of a shoe model?
It is absolutely not obvious how to solve this task. Measuring an outer surface of a shoe is pointless in most cases because there is no strict correspondence between an outer and inner shape. There is no good method even for rough measurement using a simple ruler — how to properly insert a ruler into a shoe and which points should we use to apply the ruler?
We have tried several different approaches.
The conducted research on shoe measuring methods leads us to an interesting conclusion. No matter how precise we reconstruct the 3D shape of a specific shoe model, we can’t apply this information to find the best fit, if we don’t know anything about the material from which the shoes were made and its properties. There are plenty of different features and designers’ tricks, which make a unique set of properties for each pair of shoes and have a significant impact on user experience.
For example, Nike MX-720 has a very soft sole, which instantly deforms by a human’s foot weight. So, the initial shape of this model is useless, because it differs from the shape of a shoe, worn by a real human.
Another example is Adidas Derrupt — this model has very thin and lightweight mesh material. This fabric is very flexible and can take absolutely any shape. Just as it is pointless to measure a sock, it is also pointless to try to measure this sneaker.
The soft and flexible material is not the only problem. Too tough materials also can cause some trouble. For example, Puma X-Ray 2 is a winter model and has a strong skeleton with a waterproof fabric. This means that even very tiny incompatibility of the foot shape and the shoe shape leads to the inconvenience of wearing these sneakers. A fitting method should be aware of this in order to produce good results.
To overcome this problem we decided to use the most powerful tool in our toolset — real humans :)
The scheme is actually pretty simple
- We gather representative groups of people and all sizes of the specific shoe model
- Using all available techniques (a simple ruler, The Brannock Device, our TrueDepth scanner) we measured people’s feet and did the same for a shoe
- Ask people to try all sizes of the shoe model and answer our special questions about this model
- Using advanced machine learning algorithms train an ML-model on all this data, which can answer the following question — “Given these feet measurements, which size will be the most comfortable?”
This is a very indirect way to convert feet measurements into a shoe size, but this approach has a very pleasant property — quantity turns into quality. So collecting more and more data improves our understanding of feet-shoe correspondence and allows us to make just a perfect recommendation for each specific shoe model.
Moreover, this method opens up even cooler opportunities.
What about people’s fit preferences?
The comfort of wearing shoes is not just a physical property, which you can measure with some device. It is a psychological phenomenon, a sense in our brain just like the sense of taste. Some people like it when shoes fit snugly and their feet feel every side of a model. Others don’t like too much pressure and want a more loose fit.
It is obvious that there is nothing we can do even with a high-technological measurement device. However, since we started to use a human in our process, maybe we can extract some information about fitting preferences as well?
During our measuring process, we also ask assessors about their preferences and current shoe models, which they are wearing. We didn’t really hope for success in this experiment, but the obtained results exceeded our expectations. Our final recommendation model is aware of user preferences and can apply this information for correcting measurements-only recommendations if needed.
In conclusion
As you can see, the situation with shoe sizes is a little bit complicated. The beautiful idea that there is a simple one-to-one correspondence between “length of feet” and “shoe size unit” crashes dramatically against the real world.
At Neatsy we have developed a user-friendly solution, which uses the maximum of available information from a user to overcome all these difficulties described above and recommend to him just a perfect size for each specific shoe. If you have an iPhone (especially with FaceID functionality) you can touch the future and try our application.
So, are you still sure that you really know your shoe size?
