Supplementary Material for OctNet : Learning Deep 3 D Representations at High Resolutions

In the following supplemental material we present details regarding our operations on the hybrid grid-octree data structure, additional experimental results and all details on the used network architectures. In Section 2 we provide details on the data index used to efficiently access data in the grid-octree data structure. Section 3 yields more insights into the efficient implementation of the convolution operation on octrees. We present further quantitative and qualitative results in Section 4 and in Section 5 we specify all details of the network architectures used in our experiments.