CPL-Bench Dataset

Overview

CPL-Bench introduces a new unified benchmark for cross-platform LiDAR semantic segmentation, addressing the lack of dedicated benchmarks across heterogeneous acquisition platforms. It provides town-scale MLS, ULS, and ALS point clouds from shared CARLA scenes with consistent semantics and annotations, enabling controlled cross-platform evaluation.

The dataset provides two variants:

• FullDataset — complete version for large-scale experiments (CPL-Bench/full)
• Lightweight Subset — only xyz & label (CPL-Bench/lightweight)

Dataset Structure

  CPL-Bench/
  │
  ├── full/
  │ └── ...
  │
  ├── lightweight/
  │ ├── CPL-ALS_Reduced/
  │ │ ├── Town01ALS.ply
  │ │ ├── ...
  │ │ └── Town10ALS.ply
  │ │ 
  │ ├── CPL-ULS_Reduced/
  │ │ ├── ...
  │ │ └── Town10ULS.ply
  │ │ 
  │ └── CPL-MLS_Reduced/
  │   ├── ...
  │   └── Town10MLS.ply
  │
  ├── CPL_ColorMap.xml
  │
  ├── class_mapping.yaml
  │
  └── README.md

Citation