Spatio-temporal Graph-RNN for Point Cloud Prediction
This work addresses point cloud prediction for applications like robotics or animation, but it appears incremental as it builds on existing Graph-RNN approaches with added geometric feature learning.
The paper tackles the problem of predicting future frames in point cloud sequences by introducing an end-to-end network that learns topological information as geometric features and uses Graph-RNN cells to model point dynamics. The method outperforms baselines that neglect geometry features on datasets like MINST moving digits, synthetic human motions, and JPEG dynamic bodies.
In this paper, we propose an end-to-end learning network to predict future frames in a point cloud sequence. As main novelty, an initial layer learns topological information of point clouds as geometric features, to form representative spatio-temporal neighborhoods. This module is followed by multiple Graph-RNN cells. Each cell learns points dynamics (i.e., RNN states) by processing each point jointly with the spatio-temporal neighbouring points. We tested the network performance with a MINST dataset of moving digits, a synthetic human bodies motions and JPEG dynamic bodies datasets. Simulation results demonstrate that our method outperforms baseline ones that neglect geometry features information.