Skip to content

Latest commit

 

History

History
13 lines (10 loc) · 721 Bytes

File metadata and controls

13 lines (10 loc) · 721 Bytes

Discussion over paper: "Neural LiDAR Fields for Novel View Synthesis"

Summarization

  • Highlight

    1. NFL: Neural Fields for LiDAR,一个NeRF形式的场景表达,用来解决LiDAR NVS问题
    2. 结合了NeRF的体渲染能力和LiDAR的物理采集模型,提升了生成质量
    3. 可用来增强下游任务,如point cloud registration, semantic segmentation等
  • Limitation

    1. LiDAR的物理属性与semantic feature and learning的balance还有继续优化空间
    2. Prediction of the second return mask的优化将会继续提升效果
    3. 因依赖NeRF,所以需要per-scene optimization,Generalization accross scenes和处理动态场景将是未来工作方向