Addressing Overfitting on Pointcloud Classification using Atrous XCRF

  • 2019-02-08 14:20:42
  • Hasan Asyari Arief, Ulf Geir Indahl, Geir-Harald Strand, HÃ¥vard Tveite
  • 3

Abstract

Advances in techniques for automated classification of pointcloud dataintroduce great opportunities for many new and existing applications. However,with a limited number of labeled points, automated classification by a machinelearning model is prone to overfitting and poor generalization. The presentpaper addresses this problem by inducing controlled noise (on a trained model)generated by invoking conditional random field similarity penalties usingnearby features. The method is called Atrous XCRF and works by forcing atrained model to respect the similarity penalties provided by unlabeled data.In a benchmark study carried out using the ISPRS 3D labeling dataset, ourtechnique achieves 84.97% in term of overall accuracy, and 71.05% in term of F1score. The result is on par with the current best model for the benchmarkdataset and has the highest value in term of F1 score.

 

Introduction (beta)

None

 

Conclusion (beta)

None