Object Detection Based on Transfer Learning Techniques and Transformer

Ke Xu

2025

Abstract

At present, researchers treat object detection as a hot topic especially using transformer. The goal is to categorize individual objects and determine their location by means of a bounding box. Object detection is the basis for many applications such as surveillance, image retrieval, automatic driving and face recognition. This paper proposes a newly designed transformer model to cope with the object detection task based on migration learning techniques. Specifically, the model consists of several parts: a Resnet-101 model as a backbone, an encoder with an attention mechanism, a decoder with an object query input, and a network that can fetch the output banding box. In addition, an ensemble prediction loss function for bipartite frame matching is developed. With respect to the experiment results, the paper shows it is useful to implement transfer learning technique between COCO 2007 and PASCAL VOC 2012 dataset. The Fsater Region Convolutional Neural Network (R-CNN) model was used for the model comparison. This research demonstrates the broad promise of using migration learning for object detection, enabling many downstream tasks in this area. It also lays a solid foundation for future semantic segmentation tasks using other improved converter models, which will have a real impact on applications in computer vision.

Paper Citation

in Harvard Style

Xu K. (2025). Object Detection Based on Transfer Learning Techniques and Transformer. In Proceedings of the 2nd International Conference on Data Science and Engineering - Volume 1: ICDSE; ISBN 978-989-758-765-8, SciTePress, pages 47-51. DOI: 10.5220/0013677900004670

in Bibtex Style

@conference{icdse25,
author={Ke Xu},
title={Object Detection Based on Transfer Learning Techniques and Transformer},
booktitle={Proceedings of the 2nd International Conference on Data Science and Engineering - Volume 1: ICDSE},
year={2025},
pages={47-51},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0013677900004670},
isbn={978-989-758-765-8},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 2nd International Conference on Data Science and Engineering - Volume 1: ICDSE
TI - Object Detection Based on Transfer Learning Techniques and Transformer
SN - 978-989-758-765-8
AU - Xu K.
PY - 2025
SP - 47
EP - 51
DO - 10.5220/0013677900004670
PB - SciTePress