================================================================
This code has the source code for the paper "Re-ranking Person Re-identification with k-reciprocal Encoding". Including:
- IDE baseline
- Re-ranking code
- CUHK03 new training/testing protocol
If you find this code useful in your research, please consider citing:
@article{zhong2017re,
title={Re-ranking Person Re-identification with k-reciprocal Encoding},
author={Zhong, Zhun and Zheng, Liang and Cao, Donglin and Li, Shaozi},
booktitle={CVPR},
year={2017}
}
The neighbor encoding method of our paper is inspired by the reference [2]. If you use the re-ranking code in your paper, please also cite:
@article{bai2016sparse,
title={Sparse contextual activation for efficient visual re-ranking},
author={Bai, Song and Bai, Xiang},
journal={IEEE Transactions on Image Processing},
year={2016},
publisher={IEEE}
}
================================================================
A python version of re-ranking reproduced by Hao Luo ([email protected]) is in the "python/" folder. Thanks Hao!
================================================================
The new protocol splits the CUHK03 dataset into training set and testing set similar to that of Market-1501, which consist of 767 identities and 700 identities respectively.
In testing, we randomly select one image from each camera as the query for each identity and use the rest of images to construct the gallery set. We make sure that each query identity is selected by both two cameras, so that cross-camera search can be performed.
In evaluation, true matched images captured in the same camera as the query are viewed as “junk”. Meaning that junk images is of zero influence to re-id accuracy (CMC/mAP).
The new training/testing protocol split for CUHK03 in our paper is in the "evaluation/data/CUHK03/" folder.
- cuhk03_new_protocol_config_detected.mat
- cuhk03_new_protocol_config_labeled.mat
Labeled | detected | |
---|---|---|
#Training | 7,368 | 7,365 |
#Query | 1,400 | 1,400 |
#Gallery | 5,328 | 5,332 |
- | Labeled | Labeled | detected | detected | |
---|---|---|---|---|---|
Methods | Rank@1 | mAP | Rank@1 | mAP | Reference |
BOW+XQDA | 7.93% | 7.29% | 6.36% | 6.39% | "Scalable person re-identification: a benchmark", Zheng Liang, Shen Liyue, Tian Lu, Wang Shengjin, Wang Jingdong and Tian, Qi, ICCV 2015 [project] |
PUL | - | - | 9.1% | 9.2% | "Unsupervised Person Re-identification: Clustering and Fine-tuning", Hehe Fan, Liang Zheng and Yi Yang, arXiv:1705.10444 [code] |
LOMO+XQDA | 14.8% | 13.6% | 12.8% | 11.5% | "Person Re-identification by Local Maximal Occurrence Representation and Metric Learning", Liao Shengcai, Hu Yang, Zhu Xiangyu and Li Stan Z, CVPR 2015 [project] |
IDE | 22.2% | 21.0% | 21.3% | 19.7% | "Person Re-identification: Past, Present and Future", Zheng Liang, Yi Yang, and Alexander G. Hauptmann, arXiv:1610.02984 [code] |
IDE+DaF | 27.5% | 31.5% | 26.4% | 30.0 | "Divide and Fuse: A Re-ranking Approach for Person Re-identification", Rui Yu, Zhichao Zhou, Song Bai, Xiang Bai, BMVC 2017 |
IDE+XQ.+re-ranking | 38.1% | 40.3% | 34.7% | 37.4% | "Re-ranking Person Re-identification with k-reciprocal Encoding", Zhun Zhong, Liang Zheng, Donglin Cao,Shaozi Li, CVPR 2017 [code] |
PAN | 36.9% | 35.0% | 36.3% | 34.0% | "Pedestrian Alignment Network for Person Re-identification", Zhedong Zheng, Liang Zheng, Yi Yang, arXiv:1707.00408 [code] |
DPFL | 43.0% | 40.5% | 40.7% | 37.0% | "Person Re-Identification by Deep Learning Multi-Scale Representations", Yanbei Chen, Xiatian Zhu, Shaogang Gong |
SVDNet | 40.93 | 37.83 | 41.5% | 37.26% | "SVDNet for Pedestrian Retrieval", Sun Yifan, Zheng Liang, Deng Weijian, Wang Shengjin, ICCV 2017 |
TriNet+Random Erasing | 58.14% | 53.83% | 55.50% | 50.74% | "Random Erasing Data Augmentation", Zhun Zhong, Liang Zheng, Guoliang Kang, Shaozi Li, Yi Yang, arXiv 2017 |
TriNet+Random Erasing+Re. | 63.93% | 65.05% | 64.43% | 64.75% | "Random Erasing Data Augmentation", Zhun Zhong, Liang Zheng, Guoliang Kang, Shaozi Li, Yi Yang, arXiv 2017 |
================================================================
Requirements for Caffe
and matcaffe
(see: Caffe installation instructions)
-
Build Caffe and matcaffe
cd $Re-ranking_ROOT/caffe # Now follow the Caffe installation instructions here: # http://caffe.berkeleyvision.org/installation.html make -j8 && make matcaffe
-
Download pre-computed imagenet models, Market-1501 dataset and CUHK03 dataset
Please download the pre-trained imagenet models and put it in the "data/imagenet_models" folder.
Please download Market-1501 dataset and unzip it in the "evaluation/data/Market-1501" folder.
Please download CUHK03 dataset and unzip it in the "evaluation/data/CUHK03" folder.
- Training
cd $Re-ranking_ROOT
# train IDE ResNet_50 for Market-1501
./experiments/Market-1501/train_IDE_ResNet_50.sh
# train IDE ResNet_50 for CUHK03
./experiments/CUHK03/train_IDE_ResNet_50_labeled.sh
./experiments/CUHK03/train_IDE_ResNet_50_detected.sh
- Feature Extraction
cd $Re-ranking_ROOT/evaluation
# extract feature for Market-1501
matlab Market_1501_extract_feature.m
# extract feature for CUHK03
matlab CUHK03_extract_feature.m
- Evaluation
# evaluation for Market-1501
matlab Market_1501_evaluation.m
# evaluation for CUHK03
matlab CUHK03_evaluation.m
You can download our pre-trained IDE models and IDE features, and put them in the "output" and "evaluation/feat" folder, respectively.
-
IDE models [Baiduyun] [Google drive]
-
IDE features [Baiduyun] [Google drive]
Using the above IDE models and IDE features, you can reproduce the results with our re-ranking method as follows:
- Market-1501
Methods | Rank@1 | mAP |
---|---|---|
IDE_ResNet_50 + Euclidean | 78.92% | 55.03% |
IDE_ResNet_50 + Euclidean + re-ranking | 81.44% | 70.39% |
IDE_ResNet_50 + XQDA | 77.58% | 56.06% |
IDE_ResNet_50 + XQDA + re-ranking | 80.70% | 69.98% |
For Market-1501, these results are better than those reported in our paper, since we add a dropout = 0.5 layer after pool5.
- CUHK03 under the new training/testing protocol
Labeled | Labeled | detected | detected | |
---|---|---|---|---|
Methods | Rank@1 | mAP | Rank@1 | mAP |
BOW + XQDA [1] | 7.93% | 7.29% | 6.36% | 6.39% |
BOW + XQDA + re-ranking | 8.93% | 9.94% | 8.29% | 8.81% |
LOMO + XQDA [3] | 14.8% | 13.6% | 12.8% | 11.5% |
LOMO + XQDA + re-ranking | 19.1% | 20.8% | 16.6% | 17.8% |
IDE_CaffeNet + Euclidean | 15.6% | 14.9% | 15.1% | 14.2% |
IDE_CaffeNet + Euclidean + re-ranking | 19.1% | 21.3% | 19.3% | 20.6% |
IDE_CaffeNet + XQDA | 21.9% | 20.0% | 21.1% | 19.0% |
IDE_CaffeNet + XQDA + re-ranking | 25.9% | 27.8% | 26.4% | 26.9% |
IDE_ResNet_50 + Euclidean | 22.2% | 21.0% | 21.3% | 19.7% |
IDE_ResNet_50 + Euclidean + re-ranking | 26.6% | 28.9% | 24.9% | 27.3% |
IDE_ResNet_50 + XQDA | 32.0% | 29.6% | 31.1% | 28.2% |
IDE_ResNet_50 + XQDA + re-ranking | 38.1% | 40.3% | 34.7% | 37.4% |
[1] Scalable Person Re-identification: A Benchmark. Zheng, Liang and Shen, Liyue and Tian, Lu and Wang, Shengjin and Wang, Jingdong and Tian, Qi. In ICCV 2015.
[2] Sparse contextual activation for efficient visual re-ranking. Bai, Song and Bai, Xiang. IEEE Transactions on Image Processing. 2016
[3] Person re-identification by local maximal occurrence representation and metric learning. Liao S, Hu Y, Zhu X, et al. In CVPR. 2015
If you have any questions about this code, please do not hesitate to contact us.