Authors
Michael Weinmann, Juergen Gall, and Reinhard Klein
Institute of Computer Science, University of Bonn
Portals
Summary
In this paper, we have presented an approach for creating synthetic training samples for material classification. This way, it is possible to decouple the acquisition of the material samples from the acquisition of the illumination conditions under which the material is observed. To evaluate our approach, we acquired a database of BTFs, containing 7 classes with 12 samples each, from which the training data is generated.
Abstract
To cope with the richness in appearance variation found in real-world data under natural illumination, we propose to synthesize training data capturing these variations for material classification. Using synthetic training data created from separately acquired material and illumination characteristics allows to overcome the problems of existing material databases which only include a tiny fraction of the possible real-world conditions under controlled laboratory environments. However, it is essential to utilize a representation for material appearance which preserves fine details in the reflectance behavior of the digitized materials. As BRDFs are not sufficient for many materials due to the lack of modeling mesoscopic effects, we present a high-quality BTF database with 22,801 densely measured view-light configurations including surface geometry measurements for each of the 84 measured material samples. This representation is used to generate a database of synthesized images depicting the materials under different view-light conditions with their characteristic surface geometry using image-based lighting to simulate the complexity of real-world scenarios. We demonstrate that our synthesized data allows classifying materials under complex real-world scenarios.
Contribution
- a technique for decoupling the acquisition of material samples from the environment conditions by generating synthetic training samples
- a publicly available novel BTF database of 7 material categories, each consisting of measurements of 12 different material samples, measured in a darkened lab environment with controlled illumination
- a second, novel database containing data synthesized under natural illumination which is a clear difference to other datasets which only use directional illumination or an additional single ambient illumination
- an evaluation which shows that these synthetic training samples can be used to classify materials in photographs under natural illumination conditions
Related Works
Databases; Synthetic Training Data
Overview
Based on the descriptors extracted from the synthetic training data (where the masks for the presence of materials are automatically given) we calculate a dictionary via k-means clustering. This dictionary is used to encode the descriptors per masked region via VLADs. Then, a dimensionality reduction of these VLADs is performed via PCA which is followed by an SVM-based classification.