ProxyNCA++: Revisiting and Revitalizing Proxy Neighborhood Component Analysis

This repo consists of the source code for the ProxyNCA++ paper

Make sure to download the corresponding dataset to the correct folder as specified in dataset/config.json We also include script to convert the dataset to hdf5 format.

To run the code

# CUB
CUDA_VISIBLE_DEVICES=0,1 python train.py --dataset cub  --config config/cub.json --mode train --apex --seed 0
CUDA_VISIBLE_DEVICES=0,1 python train.py --dataset cub  --config config/cub.json --mode trainval --apex --seed 0

# CARS
CUDA_VISIBLE_DEVICES=0,1 python train.py --dataset cars  --config config/cars.json --mode train --apex --seed 0
CUDA_VISIBLE_DEVICES=0,1 python train.py --dataset cars  --config config/cars.json --mode trainval --apex --seed 0

# SOP
CUDA_VISIBLE_DEVICES=0,1 python train.py --dataset sop  --config config/sop.json --mode train --apex --seed 0
CUDA_VISIBLE_DEVICES=0,1 python train.py --dataset sop  --config config/sop.json --mode trainval --apex --seed 0

# INSHOP
CUDA_VISIBLE_DEVICES=0,1 python train.py --dataset inshop  --config config/inshop.json --mode trainval --apex --seed 0

The following is the Bibtex of our paper:

@article{teh2020proxynca++,
  title={ProxyNCA++: Revisiting and Revitalizing Proxy Neighborhood Component Analysis},
  author={Teh, Eu Wern and DeVries, Terrance and Taylor, Graham W},
  journal={arXiv preprint arXiv:2004.01113},
  year={2020}
}

Acknowledgement

We want to thank dichotomies for sharing their ProxyNCA implementation (https://github.com/dichotomies/proxy-nca.). In our work, we extend their codebase structure to create ProxyNCA++.