OpenSurfaces - A Richly Annotated Catalog of Surface Appearance

One question I always ask myself when I make a new object within Blender is how should I set up its materials and textures to obtain a “believable” surface appearance (I haven’t used the term “realistic” on purpose).

After I discovered Blender 3D, two years ago, I know now how to create various surface maps (diffuse, specular/reflectance, bump, normals etc.) from a starting image, but how should they be combined to successfully create a believable material surface? Are these maps correct? How much they approximate the real material surface appearance? Which parameters and values will give better approximation of the real surface?

Of course, the trial and error approach works well now with Cycles’ preview, especially if you know exactly how the material you are working on behaves in the real world; perhaps you have that material in front of you, in your hands, and through careful observation you may guess what and how much to tweak the various elements which make up that material, but it’s not always easy and in some cases very difficult due to lack of references or material’s data.

So, having recently completed Andrew Price’s “The Architectural Academy” course I wanted to further improve my “archviz” results, so I started researching the Web for something which might help me to achieve better results.

Therefore, in case you are not yet aware of its existance, I wish to share with all of you, users and developers, what I’ve recently discovered crawling the Web, hoping it might be of interest to you.

OpenSurfaces - A Richly Annotated Catalog of Surface Appearance

Developed by Sean Bell, Paul Upchurch, Noah Snavely, Kavita Bala of the Cornell Graphics and Vision Group at the Cornell University.

As the homepage states: “OpenSurfaces is a large database of annotated surfaces created from real-world consumer photographs. Our annotation framework draws on crowdsourcing to segment surfaces from photos, and then annotate them with rich surface properties, including material, texture and contextual information.”


I suggest you to read the papers first to understand what’s behind this very interesting work and how to use the data.

Hoping that this valuable resource will be not only a useful reference to the Blender users community but also of inspiration to the brilliant developers which will have the challenging task to bring Blender’s Framework to the next level.

Good life and happy blending!

Wow, that’s clever. I guess if it would be possible to process their data to obtain cycles materials. So we could have a library that matches their database.

Nice catch. Thank you! Sometimes being able to get a good image of “How it looks” can be a huge help.

I looked at the reflectances section and I have to say, I don’t quite see how they can easily be translated into Cycles materials if there’s no indication as to what common CG material option would map to what parameter, because I thought I can make sense of some of them, but then again not.

Seems to be related to this paper: A psychophysically-based model of surface gloss perception
and this: talk - Department of Computer Science

I am glad that you guys found this information interesting. Concerning it’s usage in CG, I wrote an email to Professor Kavita Bala who managed this project asking exactly how these data can be used within a software application like Blender, but I haven’t received her reply yet. I will try again putting in cc also the other contributors to the project hoping that they might take it in consideration. We’ll see.

Thank you bashi for the links to the paper which I will read.

The interesting point about this database is that the materials have been analized within the context they were taken from and for archviz this is great.