Part-based Weighting Aggregation (PWA)

Code for our AAAI2018 paper：

Unsupervised Part-based Weighting Aggregation of Deep Convolutional Features for Image Retrieval. (paper)

Jian Xu, Cunzhao Shi, Chengzuo Qi, Chunheng Wang*, Baihua Xiao

@inproceedings{ PWA,
author = {Jian Xu and Cunzhao Shi and Chengzuo Qi and Chunheng Wang and Baihua Xiao},
title = {Unsupervised Part-Based Weighting Aggregation of Deep Convolutional Features for Image Retrieval},
conference = {AAAI Conference on Artificial Intelligence},
year = {2018}
}

NOTE:

tools:
1.The python code is based on the python data science platform Anaconda2.
2.The python code is tested on Windows by PyCharm.

data:
3.The features of convolutional layer(Pool5 layer) of VGG16 for Oxford5k and Paris6k datasets are in path "data\feature".
4.The order of part detectors are in path "data\filter_select".
5.The groundtruth for Oxford5k and Paris6k datasets are in path "data\gt_files".

code:
6.Run evaluate.py, the mAP is printed.
7.Run select_filter.py to get the order of part detectors according to variances.