As illustrated in FIG. 6, the system 100 may process distance data (e.g., K-d ordered BPS-distances 610) using a classification model 620 to generate classification probabilities 630, which indicates the object label associated with the point cloud data input to the system 100. The classification model 620 may process groups of BPS-distances block-wise according to the kd-ordering of the basis points. The classification model 620 may correspond to a network consisting of blocks of one-dimensional (1D) locally connected layers. Each of the convolutions may be followed by rectified units, dropout, and batch normalization layers.

In the example illustrated in FIG. 6, the classification model 620 includes four layers (a-d), such that a first layer a includes six blocks, a second layer b includes three blocks, a third layer c includes one block, and a fourth layer d only includes a single layer. Each block comprises a feature classification layer 622, a dropout layer 624, and a batch normalization layer 626.

The classification model 620 may output a plurality of classification probabilities 630 that indicates a likely classification for the input point cloud data. For example, if there are 100 potential classification labels, the classification probabilities 630 may include 100 separate values indicating a probability value corresponding to each class. The system 100 may identify a highest probability value and determine a corresponding classification label to associate with the input point cloud data.