🤖 AI Summary
Existing point cloud scene generation methods rely on partial scans as conditioning inputs, leading to a mismatch between training and inference, poor handling of sparsity in distant regions and occluded areas, and limited flexibility in generating scenes without LiDAR observations. To address these limitations, this work proposes a unified generation framework that dispenses with partial scans by predicting density, height, and occupancy masks in bird’s-eye view (BEV) to construct structured point sources. Furthermore, it introduces a teacher–student approximate optimal transport mechanism that learns straighter transport paths for efficient single-step point generation. The approach supports both unconditional and multi-cue conditional generation, achieving state-of-the-art Jensen–Shannon divergence (JSD) and voxel IoU on SemanticKITTI, and the best Coverage score on KITTI-360 under unconditional generation.
📝 Abstract
Existing point-based generative methods for outdoor scenes primarily focus on LiDAR-conditioned completion. During training, noisy point clouds are constructed by perturbing complete ground-truth scenes, whereas during inference, they are initialized by adding noise to duplicated partial scans. This train-inference mismatch inherits the sparsity and visibility bias of partial scans, leading to sparse distant regions and incomplete geometry in occluded areas. Moreover, the reliance on partial scans restricts generation when LiDAR observations are unavailable or replaced by layout cues. We present FPSGen, a flexible framework that constructs point sources independently of partial scans. FPSGen first predicts a bird's-eye-view (BEV) prior with density, height, and mask channels from the active cues. The density map is then sampled to form a BEV-supported point source, enabling both unconditional and conditioned initialization. A teacher-student approximate optimal transport scheme then uses teacher-predicted endpoints to learn a velocity field that induces straighter transport paths. By integrating BEV point source construction with path-straightening transport, FPSGen provides a unified framework for unconditional and flexible cue-conditioned scene generation. Extensive experiments show that FPSGen achieves state-of-the-art JSD and voxel IoU performance on SemanticKITTI completion while maintaining strong performance with a single point transport step. On KITTI-360 unconditional generation, it also achieves the best Coverage (COV) among the compared methods.