This repository contains the source code of patches/features extraction and PCA learning.
- OpenCV 3.0 or newer.
- OpenBLAS + LAPACKE (for PCA operations).
Tested on Mac OSX with GNU C/C++ compilers.
Run
$ mkdir build && cd build
$ cmake .. -DOPENBLAS_INCLUDE_DIR=</path/to/OpenBLAS/include>
$ make all
to compile the source code.
- To extract patches from an image:
$ ./main -c extract_patches -i ../IMG_1069.JPG -o ./patches -r 20
-
Now to extract the local deep features from patch images, see here.
-
To apply pca to extracted features:
$ ./main -c compute_pca -i raw_data.fvecs -o raw_data_128.fvecs -r 128 -D 4096
where -D 4096 is the dimensionality of the features, and -r 128 is the number of principal components (PCs) to be learnt.
This project only officially support GNU GPLv3 license. Please see LICENSE.
Please do not hesitate to contact me at tuan.nguyenanh@hotmail.com or t_nguyen@hal.t.u-tokyo.ac.jp.
If you use this source code, please cite the following reference:
@techreport{TuanNguyen2016,
author = {Nguyen, Tuan Anh and Duta, Ionut Cosmin and Yamasaki, Toshihiko and Aizawa, Kiyoharu},
institution = {The University of Tokyo},
title = {{Evaluation of Deep Features with PCA for Fine-grained Image Retrieval}},
year = {2016}
}