PointSSC: A Cooperative Vehicle-Infrastructure Point Cloud Benchmark for Semantic Scene Completion
Core Concepts
Semantic Scene Completion (SSC) benchmark for outdoor point cloud scenes with long-range perception and minimal occlusion.
Abstract
PointSSC introduces a benchmark for semantic scene completion using point clouds from cooperative vehicle-infrastructure views. It addresses the limitations of existing SSC models by providing a challenging testbed for real-world navigation. The dataset is developed based on V2X-Seq, offering a comprehensive dataset with spatial coverage and data volume. The model proposed for PointSSC includes a Spatial-Aware Transformer and Completion and Segmentation Cooperative Module to advance semantic point cloud completion. Experimental results show the superiority of the PointSSC model over existing methods in terms of completeness and segmentation accuracy.
PointSSCは、協力的な車両インフラストラクチャービューを使用したセマンティックシーン補完のためのベンチマークを導入します。既存のSSCモデルの制限に対処するため、実世界ナビゲーション向けの厳しい試験環境を提供します。データセットはV2X-Seqに基づいて開発され、空間カバレッジとデータ量が豊富な包括的なデータセットを提供します。PointSSC向けに提案されたモデルには、Spatial-Aware TransformerとCompletionおよびSegmentation Cooperative Moduleが含まれており、セマンティックポイントクラウド補完を進化させることができます。実験結果は、PointSSCモデルが完全性とセグメンテーション精度において既存の方法に優越性を示しています。
PointSSC
Stats
PointSSC provides a challenging testbed for semantic point cloud completion.
The dataset is derived from V2X-Seq, offering large data volume and spatial coverage.
The model includes a Spatial-Aware Transformer and Completion and Segmentation Cooperative Module.
Experimental results demonstrate the superiority of the PointSSC model over existing methods.
Quotes
"Most current SSC datasets rely on vehicle-mounted sensors, which have limited perception range."
"Infrastructure sensors possess longer range and fewer blind spots compared to vehicle sensors."
"Our model explores semantic point cloud completion for large outdoor scenes."