Flickr Material Database (FMD)
Lavanya Sharan1 Ce Liu1,2 Ruth Rosenholtz1 Edward H. Adelson1
1 Massachusetts Institute of Technology 2 Microsoft Research New England
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
| Fabric | Foliage | Glass | Leather | Metal | Paper | Plastic | Stone | Water | Wood |
| "an alternative and very difficult material dataset" |
— Forsyth & Ponce, Computer Vision: A Modern Approach, 2nd ed. |
The Flickr Material Database (FMD) consists of color photographs of surfaces belonging to one of ten common material categories: fabric, foliage, glass, leather, metal, paper, plastic, stone, water, and wood. There are 100 images in each category, 50 close-ups and 50 regular views. Each image contains surfaces belonging to a single material category in the foreground and was selected manually from approximately 50 candidates to ensure a variety of illumination conditions, compositions, colors, textures, surface shapes, material sub-types, and object associations.
FMD was constructed with the specific goal of capturing the natural range of material appearances. Consider the fabric column in the selection of images above. The images of the satin ribbon, the crocheted nylon cap , the stuffed snail toy, and the flannel bedding look very different from each other. These four sets of fabric surfaces have different material properties, are of different colors and sizes, and have distinct uses as objects. And yet, it is clear that these surfaces belong to the fabric category and not any of the others. This intentional diversity of FMD images reduces the chances that simple, low-level information (e.g., color) can be used, either by humans or by computers, to distinguish material categories.
We have used FMD to study the human perception of material categories as well as to design computer vision systems for material recognition. Details can be found in our JoV'14, IJCV'13, and CVPR'10 papers. Although FMD was originally developed to study human material perception, it has become a benchmark for material recognition in the computer vision community. State-of-the-art computer vision systems achieve <60% accuracy at identifying FMD categories (chance = 10%), whereas humans achieve 84.9%. Therefore, there is still progress to be made in designing computer vision systems that can match human performance on FMD.
Download the full database here. The zip file includes the original photographs along with masks that identify regions of interest in each image. To browse the database in high resolution, click on any of the thumbnail images above or explore per-category high resolution images here.
To cite the database, please use:
L. Sharan, R. Rosenholtz, and E. H. Adelson, "Accuracy and speed of material categorization in real-world images", Journal of Vision, vol. 14, no. 9, article 12, 2014 [BibTex]
To cite our material recognition systems, please use:
C. Liu, L. Sharan, E. H. Adelson, and R. Rosenholtz, "Exploring features in a Bayesian framework for material recognition", in Proc. of IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 239-246, 2010 [BibTex]
L. Sharan, C. Liu, R. Rosenholtz, and E. H. Adelson, "Recognizing materials using perceptually inspired features", International Journal of Computer Vision, vol. 108, no. 3, pp. 348-371, 2013 [BibTex]
This work was supported by NIH grants R01-EY019262 and R21-EY019741 and a grant from NTT Basic Research Laboratories. We thank Aseema Mohanty for help with database creation, and for discussions: Aude Oliva, Michelle R. Greene, Barbara Hidalgo-Sotelo, Molly Potter, Nancy Kanwisher, Jeremy Wolfe, Roland W. Fleming, Shin’ya Nishida, Isamu Motoyoshi, Micah K. Johnson, and Alvin Raj.