Based on our NIPS 2016 Paper: "CliqueCNN: Deep Unsupervised Exemplar Learning" by Miguel A. Bautista* , Artsiom Sanakoyeu* , Ekaterina Sutter, Björn Ommer.
https://asanakoy.github.io/cliquecnn/
- The paper can be downloaded from https://arxiv.org/abs/1608.08792
- Labels that we gathered for Olympic Sports can be found in olympic_sports_retrieval/data
- Bounding boxes for the OlympicSports dataset olympic_sports_retrieval/data/bboxes.tar.gz
- All our pretrained models for Olympic Sports dataset can be downloaded from here
- Caffe's deploy file: olympic_sports_retrieval/models/deploy.prototxt
- Evaluation script for Olympic Sports: olympic_sports_retrieval/calculate_roc_auc.py
- Baseline HOG-LDA similarity matrices for Olympic Sports: similarities_hog_lda.tar.gz (11.5 Gb)
If you find this code or data useful for your research, please cite
@inproceedings{cliquecnn2016,
title={CliqueCNN: Deep Unsupervised Exemplar Learning},
author={Bautista, Miguel A and Sanakoyeu, Artsiom and Tikhoncheva, Ekaterina and Ommer, Bj{\"o}rn},
booktitle={Proceedings of the Conference on Advances in Neural Information Processing Systems (NIPS)},
pages={3846--3854},
year={2016}
}