implementation of the data selection algorithm proposed in:
Alireza Zaeemzadeh, Mohsen Joneidi ( shared first authorship) , Nazanin Rahnavard, Mubarak Shah: Iterative Projection and Matching: Finding Structure-preserving Representatives and Its Application to Computer Vision. IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2019. link
- For a quick demo using MNIST, please run
python demo.py
. - For active learning experiments on UCF101 video dataset see here.
irlb: Truncated SVD by implicitly restarted Lanczos bidiagonalization for Numpy! code
t-SNE visualization of two classes of UCF-101 dataset and their representatives selected by IPM. (left) Decision function learned by using all the data. The goal of selection is to preserve the structure with only a few representatives. (right) Decision function learned by using representatives selected by IPM.
If you use IPM in your research, please use the following BibTeX entry.
@inproceedings{zaeemzadeh2019ipm,
title = {{Iterative Projection and Matching: Finding Structure-preserving Representatives and Its Application to Computer Vision}},
year = {2019},
booktitle = {Computer Vision and Pattern Recognition, 2019. CVPR 2019. IEEE Conference on},
author = {Zaeemzadeh, Alireza and Joneidi, Mohsen and Rahnavard, Nazanin and Shah, Mubarak}
}
UCF Center for Research in Computer Vision (CRCV)