BNM

code release for "Towards Discriminability and Diversity: Batch Nuclear-norm Maximization under Label Insufficient Situations" ( CVPR2020 oral)

One-sentence description

We prove in the paper that Batch Nuclear-norm Maximization (BNM) could ensure the prediction discriminability and diversity, which is an effective method under label insufficient situations.

Application

Operation approach

Assume X is the prediction matrix. We could calculate BNM loss in both Pytorch and Tensorflow, as follows:

-Pytorch

Direct calculation (Since there remains direct approach for nuclear-norm)

L_BNM = -torch.norm(X,'nuc')

Calculation by SVD

U,S,V = torch.svd(X)
L_BNM = -torch.sum(S)

-Tensorflow

L_BNM = -tf.reduce_sum(tf.svd(X,compute_uv=False))

Tasks

We apply BNM to domain adaptation (DA) in DA, unsupervised open domain recognition (UODR) in UODR and semi-supervised learning (SSL) in SSL.

Training instructions for DA, UODR and SSL are in the README.md in DA, UODR and SSL respectively.

Citation

If you use this code for your research, please consider citing:

@inproceedings{cui2020bnm,
  title={Towards Discriminability and Diversity: Batch Nuclear-norm Maximization under Label Insufficient Situations},
  author={Cui, Shuhao and Wang, Shuhui and Zhuo, Junbao and Li, Liang and Huang, Qingming and Tian Qi},
  booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},
  year={2020}
}

Contact

If you have any problem about our code, feel free to contact

hassassin1621@gmail.com

or describe your problem in Issues.