Our Scan Processing Pipeline: How we Source Biomes Everywhere in the World

How the Substance team collects, organizes, reworks, optimizes, and makes your scans procedural.

There is no terrain, no cliff, no mountain, too hard to climb for you! Today, we take you for a hike behind the curtain to discover parts of the process behind the creation of our scans collection.

It’s not just the largest collection of scanned materials ever published on Substance Source: to process it, we have built a complete asset production model. Our goal was to deliver high-quality, ready-to-use content, standardized across all types of surfaces in an almost automatic way. Let’s see how that works, from the capture to the delivery.

Step 1: Building scalability and consistency

Artwork by Paul Parneix

Historically, “artisanal” productions are the traditional approach to deliver scan-based materials. Highly skilled experts paved the way by handling every step of the production process. But for our purposes, this wasn’t going to work: we wanted to be able to handle a lot of data, in a shorter time, and most importantly, without compromising on quality.

Quality isn’t a problem when an expert travels the world, scouting for interesting patches to capture, post-processing the data sets — often manually — and packaging the end material for final delivery.

But there are obvious limitations.

First, the quantity of content is limited to what a single person can capture. Depending on travel possibilities, that also means that the terrain captured will be restricted to a small geographic zone.

Secondly, each expert tends to have their own technique to process the data and generate the content, and that creates strong variations in the standards of quality.

That was not going to suit our purposes.

Armed with patience and a bit of single-mindedness, we set to design a model that would allow us to produce at an industrial quantity with a homogeneous quality.

We separated the production steps on a model similar to a product assembly line, leading to important time savings, since most of the processing steps ended up being automated.

The result is a flexible production pipeline that enables to source surfaces from multiple remote locations simultaneously around the globe and process them homogeneously, with as few manual operations as possible.

Step 2: Scanning the world

Artwork by Maximilien Vert

Scouting and capture

Boots on the ground and elbow grease! This is a step you can’t automate: only scouting the land helps you find the right surfaces to capture.

The world is vast. Focusing on what’s around oneself won’t be representative of the incredibly diversity our planet offers. We had to find a way to access biomes in all parts of the world and capture a qualitative data set without carrying heavy machinery.

We deployed a team of partners, trained photographers who trekked in almost every continent — armed with only light cameras! Ultra-portable gear means accessing remote places quickly. It sounds mad, but it worked, and we were able to simultaneously run campaigns of capture in North and South America, Europe, India and New Zealand!

They captured a wild diversity of biomes and seasonal conditions. Some may think that grass is grass and the same green everywhere, but the richness of our planet tells otherwise.

Roads, streets and wall of cities and villages

France

USA

USA

Temperate, mediterranean and tropical forests

Hungary

Arid and desert sands and rockery

Mountain rocky trails and river beds, rock formations of several mineral composition

Sands and pebbles of many shores

And this is just the beginning: this collection represents only what was captured over a few months. As of today, we’re still capturing more places.

Unifying the scans library

Earth is a place of diversity. But diversity does not mean variable standards of quality. Consistency is critical when we produce assets that are sharing the same technical conventions.

This process let us generate high resolution maps, up to 8K, for every asset. We received surface patch captures from 2m2 to 2.5m2 in average, but they were measured by the photographers, so this helped us harmonize them.

But harmonization also requires to check that the capture matches the requirements, so we’ve designed an automated processing line.

It ingests the captures sent from all over the globe to a central service. Said service guarantees an homogeneous treatment regardless of the operator, type of surface, and dataset specificities.

Our objective, simplifying a highly technical complex process mastered by a few experts to everyone owning a camera thanks to Substance and with a little help from Alchemist’s machine learning scan processing workflow.

Tileable surfaces

When you’re working on a large scale environment, we all know that size matters! Making a scan tileable is equivalent to making it useable. Here, too, we’ve developed specific tiling tools with Substance to simplify scan post-processing operations.

Traditionally, scaling up the size of the data set requires an increase of human ressources down the line. Artists will have to manually do — or at least control — these operations to deliver seamless surface stitching.

Tiling woods, grounds, sands, or walls materials require that the artist uses a different method each time. Every method has its own tricks and to achieve a clean result, the artist may need from a few minutes to almost an hour. And the most complex materials to tile are not the ones you think!

But we were working on an industrial scale, and we could not reliably enclose 300 artists in a barn. At that scale, designing custom flexible tools to optimize operations was an absolute necessity.

And it worked! We managed to overcome the task of tiling 720+ scans with a handful of artists in less than 3 weeks (and that wasn’t even our only project at that time ;).

Making scan-based materials parametric

The job does not end at tiling. Substance Source is a library of parametric materials and that means that every assets on the catalog comes with exposed parameters. This is something we’re passionate about: making quality materials is not enough, we want to empower users to customize their material.

Scan-based materials are no exception! The collection combines the best of both worlds and biomes. You have the power to modify selectively the colors and roughness of the materials. This is now part of the pipeline: it automatically outputs extra maps like the roughness and exposes the basic parameters.

Artwork by Maximilien Vert

Presets, and most importantly color variations, are where the expertise of material artists come into play. Of course, our team operated verifications on every single scanned material we processed, but they had an important art direction role. The team selected colors and made artistic choices to provide you with presets that made sense and would help you get the most of each material.

All in all, there are 2880 presets. For 720 materials.

Step 3: Blending streams

Play

The 2021 collection is expanding the diversity of content available to for you to mix, enhance and customize. You can, for instance, use Substance Alchemist and create an even richer combination of soils and surfaces!

It all comes down to a simple formula (we like maths):

materials + atlases + filters = endless combinations

To learn more about this, check out our “Biomes” collection. In the meantime, take a peek outside, we might be scanning near you right now…