Addressing the Sim2Real Gap in Robotic 3D Object Classification
This addresses the problem of transferring classification from artificial to real data for robotics, but it is incremental as it builds on existing methods.
The paper tackles the sim2real gap in robotic 3D object classification by training on CAD models (ModelNet) and evaluating on reconstructed objects (ScanNet), showing that standard methods perform poorly and introducing a method that improves upon the baseline.
Object classification with 3D data is an essential component of any scene understanding method. It has gained significant interest in a variety of communities, most notably in robotics and computer graphics. While the advent of deep learning has progressed the field of 3D object classification, most work using this data type are solely evaluated on CAD model datasets. Consequently, current work does not address the discrepancies existing between real and artificial data. In this work, we examine this gap in a robotic context by specifically addressing the problem of classification when transferring from artificial CAD models to real reconstructed objects. This is performed by training on ModelNet (CAD models) and evaluating on ScanNet (reconstructed objects). We show that standard methods do not perform well in this task. We thus introduce a method that carefully samples object parts that are reproducible under various transformations and hence robust. Using graph convolution to classify the composed graph of parts, our method significantly improves upon the baseline.