From Simulation to Reality: Transfer Learning for Automating Pseudo‐Labeling of Real and Infrared Imagery

Training a convolutional neural network (CNN) for real‐world applications is challenging due to the requirement of high‐quality labeled imagery. This study employs pseudo‐labeling and transfer learning, built upon a 6D pose estimation framework. A CNN trained on synthetic images predicts bounding bo...

Full description

Saved in:
Bibliographic Details
Main Authors: Jeffrey Choate, Derek Worth, Scott L. Nykl, Clark Taylor, Brett Borghetti, Christine Schubert Kabban, Ryan Raettig
Format: Article
Language:English
Published: Wiley 2025-05-01
Series:Advanced Intelligent Systems
Subjects:
Online Access:https://doi.org/10.1002/aisy.202400575
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Training a convolutional neural network (CNN) for real‐world applications is challenging due to the requirement of high‐quality labeled imagery. This study employs pseudo‐labeling and transfer learning, built upon a 6D pose estimation framework. A CNN trained on synthetic images predicts bounding boxes (bbox) for an object's components in a real image. With as few as four bbox predictions, the framework solves for the object's pose relative to the camera and reprojects bboxes for all components onto that image. The pose and reprojections allow filtering of bad predictions, a common issue in pseudo‐labeling. Thereby, enabling automated labeling of large datasets with minimal human intervention. Tested on color and long‐wave infrared imagery captured during December 2023 flight tests, this process demonstrates increased predictions, enhanced performance across situations, reduced reprojection error, and stabilized pose predictions. This technique is significant as it enables labeling of real‐world imagery without expensive truth systems, requiring only a camera. It supports learning and labeling of previously captured imagery without known camera calibrations, facilitating labeled data creation for impractical‐to‐simulate sensors. Ultimately, this transfer learning approach provides a low‐cost and precise method for creating CNNs trained on operationally relevant data, previously unattainable by the everyday user.
ISSN:2640-4567