Object Categorization

Object recognition has reached a level where we can identify a large number of previously seen and known objects. However, the more challenging and important task of categorizing previously unseen objects remains largely unsolved. Traditionally, contour and shape based methods are regarded most adequate for handling the generalization requirements needed for this task. Appearance based methods, on the other hand, have been successful in object identification and detection scenarios. Today little work is done to systematically compare existing methods and characterize their relative capabilities for categorizing objects.

To address these issues, we have built a novel database specifically tailored to the task of object categorization. It contains 80 objects from 8 categories. Each object is represented by 41 views spaced evenly over the upper viewing hemisphere. This allows to analyze the performance of different recognition methods not only from a 1D circle or a few canonical viewpoints, but from multiple viewing positions. For each image a high-quality figure-ground segmentation mask is provided. This makes it possible to compare both appearance and contour based methods in the idealized setting of perfect segmentation. Even though any comparison on a particular database has its limitations, we strongly believe that databases such as the one we propose, as well as the comparison of different methods are important steps to enable progress in the area of object categorization. The database is made publicly available and other authors are invited to run and report experiments. More on the database...

We use this database to compare different methods for object categorization. In particular, we want to address the question of what the role of color, texture, and shape is for this task. For this reason, we have analyzed the performance of several state-of-the-art appearance- and contour-based recognition methods on the database categories. Interestingly, the best categorization result is obtained by an appropriate combination of multiple methods. A detailed description of the experiments and results can be seen here.