/SNCA_CE

codes for TGRS paper "Deep Metric Learning based on Scalable Neighborhood Components for Remote Sensing Scene Characterization"

Primary LanguagePython

Deep Metric Learning based on Scalable Neighborhood Components for Remote Sensing Scene Characterization

Jian Kang, Ruben Fernandez-Beltran, Zhen Ye, Xiaohua Tong, Pedram Ghamisi, Antonio Plaza.

This repo contains the codes for the TGRS paper: Deep Metric Learning based on Scalable Neighborhood Components for Remote Sensing Scene Characterization. we propose a new deep metric learning approach, which overcomes the limitation on the class discrimination by means of two different components: 1) scalable neighborhood component analysis (SNCA), which aims at discovering the neighborhood structure in the metric space; and 2) the cross entropy loss, which aims at preserving the class discrimination capability based on the learned class prototypes. Moreover, in order to preserve feature consistency among all the mini-batches during training, a novel optimization mechanism based on momentum update is introduced for minimizing the proposed loss. Some codes are modified from SNCA and CMC.

alt text

drawing

Usage

train_SNCA_CE/main.py is the script of the proposed SNCA-CE (MB) method for training and validation. train_SNCA_CE_MC/main.py is the script of the proposed SNCA-CE (MU) method for training and validation.

Citation

@article{kang2020deepmetric,
  title={{Deep Metric Learning based on Scalable Neighborhood Components for Remote Sensing Scene Characterization}},
  author={Kang, Jian and Fernandez-Beltran, Ruben and Ye, Zhen and Tong, Xiaohua and Ghamisi, Pedram and Plaza, Antonio},
  journal={IEEE Transactions on Geoscience and Remote Sensing},
  year={2020},
  note={DOI:10.1109/TGRS.2020.2991657}
  publisher={IEEE}
}

References

[1] Wu, Zhirong, Alexei A. Efros, and Stella X. Yu. "Improving generalization via scalable neighborhood component analysis." Proceedings of the European Conference on Computer Vision (ECCV). 2018.

[2] He, Kaiming, et al. "Momentum contrast for unsupervised visual representation learning." arXiv preprint arXiv:1911.05722 (2019).