We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:


Current browse context:


Change to browse by:


References & Citations

DBLP - CS Bibliography


(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Computer Vision and Pattern Recognition

Title: Supervised and Contrastive Self-Supervised In-Domain Representation Learning for Dense Prediction Problems in Remote Sensing

Abstract: In recent years Convolutional neural networks (CNN) have made significant progress in computer vision. These advancements have been applied to other areas, such as remote sensing and have shown satisfactory results. However, the lack of large labeled datasets and the inherent complexity of remote sensing problems have made it difficult to train deep CNNs for dense prediction problems. To solve this issue, ImageNet pretrained weights have been used as a starting point in various dense predictions tasks. Although this type of transfer learning has led to improvements, the domain difference between natural and remote sensing images has also limited the performance of deep CNNs. On the other hand, self-supervised learning methods for learning visual representations from large unlabeled images have grown substantially over the past two years. Accordingly, in this paper we have explored the effectiveness of in-domain representations in both supervised and self-supervised forms to solve the domain difference between remote sensing and the ImageNet dataset. The obtained weights from remote sensing images are utilized as initial weights for solving semantic segmentation and object detection tasks and state-of-the-art results are obtained. For self-supervised pre-training, we have utilized the SimSiam algorithm as it is simple and does not need huge computational resources. One of the most influential factors in acquiring general visual representations from remote sensing images is the pre-training dataset. To examine the effect of the pre-training dataset, equal-sized remote sensing datasets are used for pre-training. Our results have demonstrated that using datasets with a high spatial resolution for self-supervised representation learning leads to high performance in downstream tasks.
Subjects: Computer Vision and Pattern Recognition (cs.CV)
Cite as: arXiv:2301.12541 [cs.CV]
  (or arXiv:2301.12541v1 [cs.CV] for this version)

Submission history

From: Ali Ghanbarzade [view email]
[v1] Sun, 29 Jan 2023 20:56:51 GMT (5076kb)

Link back to: arXiv, form interface, contact.