Method for automatically generating a labeling instruction for labeling an image and system for executing the method

Automated generation of labeling instructions using large visual language models addresses the inefficiencies of manual annotation, enhancing accuracy and enabling real-time lane detection in driver assistance systems.

DE102024123825B4Active Publication Date: 2026-05-28CARIAD SE +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
DE102024123825
Authority / Receiving Office
DE · DE
Patent Type
Patents
Current Assignee / Owner
Filing Date
2024-08-21
Publication Date
2026-05-28
Estimated Expiration
2044-08-21

AI Technical Summary

Technical Problem

Current manual annotation processes for real-time data in machine learning, particularly in the automotive industry, are time-consuming and prone to errors, necessitating a more efficient and accurate automated annotation method for applications like lane recognition in driver assistance systems.

Method used

A method utilizing large machine-based visual language models, such as OpenAI CLIP and ALIGN, to automatically generate labeling instructions by training on a large dataset of images with associated labels, and iteratively refining these instructions using reconstruction loss to improve accuracy, enabling real-time annotation.

Benefits of technology

This approach reduces manual labor, enhances annotation accuracy, and enables real-time lane detection and guidance in driver assistance systems, improving data processing efficiency and reducing costs.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 00000000_0000_ABST
    Figure 00000000_0000_ABST
Patent Text Reader

Abstract

The invention relates to a method for automatically generating a labeling instruction (11) for at least one image (1) using a system comprising an instruction model and a labeling model (11, 13). In step a), images (1) with manual labels (5) are passed to the instruction model (9). In step b), the instruction model (9) generates labeling instructions (11) for the images (1). In step c), the labeling model (13) inputs these labeling instructions (11) and automatically generates labels for the images (1). In step d), a reconstruction loss (19) quantifies a discrepancy between the manual (5) and automatic labels. In step e), steps a) to d) are repeated until a certain number of training loops or a defined reconstruction loss value is reached. If a condition from step e) is met, in step b)2) the instruction model (9) was applied to new images (1) without labels.
Need to check novelty before this filing date? Find Prior Art