Virtu Tech Solutions present a new framework for 3D model retrieval based on the assumption that models belonging to the same shape class share the same salient features. The main issue is learning these features. We propose an algorithm for computing these features and their corresponding saliency value. At the learning stage, a large set of features are extracted from every model and a boosting algorithm is applied to learn the classification function in the feature space. AdaBoost learns a classifier that relies on a small subset of the features with the mean of weak classifiers. Moreover it assigns weights to the selected features, that we interpret as a measure of the feature saliency within the class, providing an efficient way for feature selection and combination. Our experiments using the LightField (LFD) descriptors and the Princeton Shape Benchmark show significant improvement in the retrieval performance and computation efficiency. We show also that the proposed framework can be applied to the problem of best view selection.
