Scholarlink

In this paper, we present a novel method for validating the placement of flooring cutouts in maps representing specific collections of flooring patterns. Due to the unique and often limited editions of these collections, traditional classification methods fall short. To address this, we propose a Siamese Convolutional Neural Network approach utilizing triplet loss to learn similarity measures between flooring patterns. By leveraging reference copies, validated by multiple employees, we aim to accurately validate the equipping of new flooring patterns. For pre-training, we utilize the material recognition in the wild dataset, focusing on class segmentation and careful extraction of non-overlapping, adequately sized pattern crops. Our method demonstrates promising results in maintaining high accuracy in validating new and unique flooring pattern classes.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Siamese CNN Approach for Validating Flooring Cutouts in Map Collections Using Triplet Loss and a Novel Pattern Crop Dataset

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Nico Rabethge,
  • Franz Kummert

摘要

In this paper, we present a novel method for validating the placement of flooring cutouts in maps representing specific collections of flooring patterns. Due to the unique and often limited editions of these collections, traditional classification methods fall short. To address this, we propose a Siamese Convolutional Neural Network approach utilizing triplet loss to learn similarity measures between flooring patterns. By leveraging reference copies, validated by multiple employees, we aim to accurately validate the equipping of new flooring patterns. For pre-training, we utilize the material recognition in the wild dataset, focusing on class segmentation and careful extraction of non-overlapping, adequately sized pattern crops. Our method demonstrates promising results in maintaining high accuracy in validating new and unique flooring pattern classes.