Fast Voxel Query

Fast Voxel Query is a module used in the Voxel Transformer 3D object detection model implementation of self-attention, specifically Local and Dilated Attention. For each querying index $v_{i}$, an attending voxel index $v_{j}$ is determined by Local and Dilated Attention. Then we can lookup the non-empty index $j$ in the hash table with hashed $v_{j}$ as the key. Finally, the non-empty index $j$ is used to gather the attending feature $f_{j}$ from $\mathcal{F}$ for multi-head attention.