Supplementary material for “3D point cloud semantic segmentation through functional data analysis”

Supplementary codes and data used in the paper “3D point cloud semantic segmentation through functional data analysis”.

Please cite this paper as:

@article{oviedo2023PointCloud,
  title={3D point cloud semantic segmentation through
functional data analysis},
  author={Oviedo de-la Fuente, Manuel, Cabo, Carlos, Roca-Pardiñas, Javier, Loudermilk, Lousie and Ordóñez Celestino},Accepted in JABES
  year={2023}
}

Installation

In order to use paper implementation and run all files (numerical and real example), the following prerequisites are needed:

1. fda.usc package

Our proposed method is available in the fda.usc package (devel version) through the functions classif.gsam.vs() and classif.ML.vs().

2. PointNet model

PointNet was improved using a hierarchically approach called PointNet++ [9].

Referencia a github

Qi R, Yi l, Su H, Guibas LJ (2017) “PointNet++: Deep hierarchical feature learning on point sets in a metric space”, Proc. Adv. Neural Inf. Process. Syst. 30 (NIPS), pp. 5105-5114

3. DGCNN model

DGCNN is a deep learning (DL) model for point cloud classification and semantic segmentation

Referencia a github

Wang Y, Yongbin S, Ziwei L, Sarma S Bronstein M, Solomo J (2018) Dynamic Graph CNN for Learning on Point Clouds. ACM Transactions on Graphics 38 2018 doi = 10.1145/3326362

Real Data Application: 3D cloud point

Our case study example is 3D cloud point.

library(fda.usc.devel)
load(file="./data/ldf.train.RData")
class(ldf.train)

## [1] "ldata" "list"

n <- nrow(ldf.train$df)
# Response
table(ldf.train$df$class)

## 
## branch+leaf        bush       grass       trunk 
##        1818        1948        1711        1770

labels <- c("trunk","grass","branch+leaf","bush")
#levEn <- levels(ldf.train$df$class)
# predictor variables (multivariate)
predictors <- names(ldf.train$df[-c(1,2,3,5,6,831)])
# linear correlation between predictors
# dfcor <- cor(ldf.train$df[,c(predictors)])

Locations

tab <- table(ldf.train$df$parcela)
iparcela <- c(0,2,6)
sum(tab[-iparcela]) # 100k,250k,500k

## [1] 5395

prop.table(tab)

## 
##         4         5         6         7 
## 0.2667311 0.2555540 0.3118532 0.1658617

ii <-1:400000
plot(ldf.train$df[ii,1:2],col=ldf.train$df$parcela[ii]+1,asp=T)
ii <- ldf.train$df$parcela %in% c(0,2,6) 
points(ldf.train$df[ii,1:2],col=1)

Functional mean curves for Planarity and Sphericity features (colored by class label).

This repository is under development, codes with examples will be included soon