Abstract:Detection pre-training methods for the DETR series detector have been extensively studied in natural scenes, e.g., DETReg. However, the detection pre-training remains unexplored in remote sensing scenes. In existing pre-training methods, alignment between object embeddings extracted from a pre-trained backbone and detector features is significant. However, due to differences in feature extraction methods, a pronounced feature discrepancy still exists and hinders the pre-training performance. The remote sensing images with complex environments and more densely distributed objects exacerbate the discrepancy. In this work, we propose a novel Mutually optimizing pre-training framework for remote sensing object Detection, dubbed as MutDet. In MutDet, we propose a systemic solution against this challenge. Firstly, we propose a mutual enhancement module, which fuses the object embeddings and detector features bidirectionally in the last encoder layer, enhancing their information interaction.Secondly, contrastive alignment loss is employed to guide this alignment process softly and simultaneously enhances detector features' discriminativity. Finally, we design an auxiliary siamese head to mitigate the task gap arising from the introduction of enhancement module. Comprehensive experiments on various settings show new state-of-the-art transfer performance. The improvement is particularly pronounced when data quantity is limited. When using 10% of the DIOR-R data, MutDet improves DetReg by 6.1% in AP50. Codes and models are available at: <a class="link-external link-https" href="https://github.com/floatingstarZ/MutDet" rel="external noopener nofollow">this https URL</a>.

Deeply Unsupervised Patch Re-Identification for Pre-training Object Detectors

Unleashing Potential of Unsupervised Pre-Training with Intra-Identity Regularization for Person Re-Identification

Unsupervised Object Detection Pretraining with Joint Object Priors Generation and Detector Learning

Unsupervised Pre-Training for Detection Transformers

UP-DETR: Unsupervised Pre-training for Object Detection with Transformers

Person Re-identification Based on Transform Algorithm

Unsupervised Pre-training for Person Re-identification

DETReg: Unsupervised Pretraining with Region Priors for Object Detection

PatchDetector: Pluggable and Non-Intrusive Patch for Small Object Detection

Self-Training With Progressive Representation Enhancement for Unsupervised Cross-Domain Person Re-Identification

Label-efficient object detection via region proposal network pre-training

Unsupervised Object Discovery and Co-Localization by Deep Descriptor Transforming

Deep Transfer Learning for Person Re-Identification

ProposalContrast: Unsupervised Pre-training for LiDAR-based 3D Object Detection

MutDet: Mutually Optimizing Pre-training for Remote Sensing Object Detection

Aligned Unsupervised Pretraining of Object Detectors with Self-training

Large-Scale Pre-training for Person Re-identification with Noisy Labels

TransReID: Transformer-based Object Re-Identification

Unified pre-training with pseudo infrared images for visible-infrared person re-identification

Multi-patch Feature Pyramid Network for Weakly Supervised Object Detection in Optical Remote Sensing Images

From Patches to Objects: Exploiting Spatial Reasoning for Better Visual Representations