Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

GPT-based Textile Pilling Classification Using 3D Point Cloud Data

Yu Lu, YuYu Chen, Gang Zhou, Zhenghua Lan

TL;DR

The paper addresses textile pilling grading by leveraging 3D point cloud data through a new public benchmark, TextileNet8. It introduces PointGPT+NN, a fusion of GPT-based point cloud analysis with global features from Point-NN, to incorporate holistic 3D information into the classification pipeline. Empirical results show PointGPT+NN achieving an overall accuracy of 91.8% and mean class accuracy of 92.2% on TextileNet8, with competitive performance on ModelNet40, underscoring the method's generalizability. The work provides a valuable dataset and demonstrates that integrating global 3D representations with transformer-based models can significantly improve textile pilling classification, offering practical benefits for quality control in textile production.

Abstract

Textile pilling assessment is critical for textile quality control. We collect thousands of 3D point cloud images in the actual test environment of textiles and organize and label them as TextileNet8 dataset. To the best of our knowledge, it is the first publicly available eight-categories 3D point cloud dataset in the field of textile pilling assessment. Based on PointGPT, the GPT-like big model of point cloud analysis, we incorporate the global features of the input point cloud extracted from the non-parametric network into it, thus proposing the PointGPT+NN model. Using TextileNet8 as a benchmark, the experimental results show that the proposed PointGPT+NN model achieves an overall accuracy (OA) of 91.8% and a mean per-class accuracy (mAcc) of 92.2%. Test results on other publicly available datasets also validate the competitive performance of the proposed PointGPT+NN model. The proposed TextileNet8 dataset will be publicly available.

GPT-based Textile Pilling Classification Using 3D Point Cloud Data

TL;DR

The paper addresses textile pilling grading by leveraging 3D point cloud data through a new public benchmark, TextileNet8. It introduces PointGPT+NN, a fusion of GPT-based point cloud analysis with global features from Point-NN, to incorporate holistic 3D information into the classification pipeline. Empirical results show PointGPT+NN achieving an overall accuracy of 91.8% and mean class accuracy of 92.2% on TextileNet8, with competitive performance on ModelNet40, underscoring the method's generalizability. The work provides a valuable dataset and demonstrates that integrating global 3D representations with transformer-based models can significantly improve textile pilling classification, offering practical benefits for quality control in textile production.

Abstract

Textile pilling assessment is critical for textile quality control. We collect thousands of 3D point cloud images in the actual test environment of textiles and organize and label them as TextileNet8 dataset. To the best of our knowledge, it is the first publicly available eight-categories 3D point cloud dataset in the field of textile pilling assessment. Based on PointGPT, the GPT-like big model of point cloud analysis, we incorporate the global features of the input point cloud extracted from the non-parametric network into it, thus proposing the PointGPT+NN model. Using TextileNet8 as a benchmark, the experimental results show that the proposed PointGPT+NN model achieves an overall accuracy (OA) of 91.8% and a mean per-class accuracy (mAcc) of 92.2%. Test results on other publicly available datasets also validate the competitive performance of the proposed PointGPT+NN model. The proposed TextileNet8 dataset will be publicly available.
Paper Structure (16 sections, 4 equations, 2 figures, 6 tables)

This paper contains 16 sections, 4 equations, 2 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Textile Pilling Evaluation
  4. 3D Point Cloud Classification
  5. Methodology
  6. Generative Pre-training Transformer(GPT) for Point Cloud
  7. Construction of PointGPT+NN
  8. Dataset
  9. Data Acquisition and Annotation
  10. Statistics of TextileNet8
  11. Experiments
  12. Experimental Settings
  13. Result Analysis
  14. Ablation Analysis
  15. Preformance on Other Dataset
  16. ...and 1 more sections

Figures (2)

  • Figure 1: Sample images from the collected point cloud data of polyester. Best viewed in color.
  • Figure 2: Overall architecture of the proposed PointGPT+NN. The input point cloud image is divided into multiple point patches, which are then sorted and arranged in an ordered sequence. The absolute position encoding (APE) is also generated in this stage. The arranged point patches are then put into a PointNet network for token embedding. Meanwhile, the input point cloud image is fed into a Point-NN network to do feature embedding on a complete point cloud image. $c$ in the Point-NN network denotes the stage number of the multi-stage hierarchy. The features generated by the PointNet and the Point-NN are then fused by element-wise addition. The fused features and the APE are then put into the extractor, while a dual mask strategy is applied to generate the logit classification results. Best viewed in color.