Hardness-Aware Deep Metric Learning

Implementation of Hardness-Aware Deep Metric Learning (CVPR 2019 Oral) in Tensorflow.

HDML: Hardness-Aware Deep Metric Learning

Work in progress.

Please use the citation provided below if it is useful to your research:

Wenzhao Zheng, Zhaodong Chen, Jiwen Lu, and Jie Zhou, Hardness-Aware Deep Metric Learning, arXiv, abs/1903.05503, 2019.

@article{zheng2019hardness,
  title={Hardness-Aware Deep Metric Learning},
  author={Zheng, Wenzhao and Chen, Zhaodong and Lu, Jiwen and Zhou, Jie},
  journal={arXiv preprint arXiv:1903.05503},
  year={2019}
}

Dependencies

pip install tensorflow==1.10.0

Dataset

Stanford Cars Dataset (Cars196)

-- Download from (https://ai.stanford.edu/~jkrause/cars/car_dataset.html) or use datasets/cars196_downloader.py.

-- Convert to hdf5 file using cars196_converter.py.

-- Put it in datasets/data/cars196/cars196.hdf5.

Pretrained model

GoogleNet V1 pretrained model can be downloaded from (https://github.com/Wei2624/Feature_Embed_GoogLeNet)

Usage

For Cars196 dataset:

python main_npair.py --dataSet='cars196' --batch_size=128 --Regular_factor=5e-3 --init_learning_rate=7e-5 --load_formalVal=False --embedding_size=128 --loss_l2_reg=3e-3 --init_batch_per_epoch=500 --batch_per_epoch=64 --max_steps=8000 --beta=1e+4 --lr_gen=1e-2 --num_class=99 --_lambda=0.5 --s_lr=1e-3

Code Reference

deep_metric_learning (https://github.com/ronekko/deep_metric_learning) by ronekko for dataset codes.

lonl/HDML