2017-03-22 18:48:01 +08:00
# Re-ranking Person Re-identification with k-reciprocal Encoding
================================================================
2017-05-05 12:59:18 +08:00
### This code has the source code for the paper "Re-ranking Person Re-identification with k-reciprocal Encoding". Including:
1. IDE baseline
2. Re-ranking code
2017-05-05 13:01:03 +08:00
3. CUHK03 new training/testing protocol
2017-03-22 18:48:01 +08:00
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}
}
2017-03-24 16:54:39 +08:00
2017-04-28 22:23:56 +08:00
The neighbor encoding method of our paper is inspired by the reference [2]. For more details of the application on image retrieval please refer to:
@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}
}
2017-03-24 17:08:21 +08:00
================================================================
2017-03-24 16:54:39 +08:00
## The new training/testing protocol for CUHK03
2017-03-24 17:08:21 +08:00
2017-05-05 16:09:37 +08:00
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).
2017-03-24 16:54:39 +08:00
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
2017-03-22 18:48:01 +08:00
2017-05-05 16:15:28 +08:00
| | Labeled | detected|
| -------| ----- | ---- |
|#Training | 7,368 | 7,365|
|#Query | 1,400 | 1,400|
|#Gallery | 5,328% | 5,332|
2017-03-24 16:55:59 +08:00
================================================================
2017-05-05 13:06:18 +08:00
## IDE Baseline + Re-ranking
2017-03-24 17:08:21 +08:00
2017-03-22 18:48:01 +08:00
### Requirements: Caffe
Requirements for `Caffe` and `matcaffe` (see: [Caffe installation instructions ](http://caffe.berkeleyvision.org/installation.html ))
### Installation
1. Build Caffe and matcaffe
```Shell
cd $Re-ranking_ROOT/caffe
# Now follow the Caffe installation instructions here:
# http://caffe.berkeleyvision.org/installation.html
make -j8 & & make matcaffe
```
2. Download pre-computed imagenet models, Market-1501 dataset and CUHK03 dataset
```Shell
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.
```
- [Pre-trained imagenet models ](https://pan.baidu.com/s/1o7YZT8Y )
- [Market-1501 ](https://pan.baidu.com/s/1ntIi2Op )
- CUHK03 [[Baiduyun]](https://pan.baidu.com/s/1o8txURK) [[Google drive]](https://drive.google.com/open?id=0B7TOZKXmIjU3OUhfd3BPaVRHZVE)
### Training and testing IDE model
1. Training
```Shell
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
```
2. Feature Extraction
```Shell
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
```
2017-03-22 18:50:24 +08:00
3. Evaluation
2017-03-22 18:48:01 +08:00
```Shell
# evaluation for Market-1501
matlab Market_1501_evaluation.m
# evaluation for CUHK03
matlab CUHK03_evaluation.m
```
### Results
2017-05-05 13:51:28 +08:00
You can download our pre-trained IDE models and IDE features, and put them in the "output" and "evaluation/feat" folder, respectively.
2017-03-22 18:48:01 +08:00
- IDE models [[Baiduyun]](https://pan.baidu.com/s/1jHVj2C2) [[Google drive]](https://drive.google.com/open?id=0B7TOZKXmIjU3ZTNsWGt3azcxUUU)
- IDE features [[Baiduyun]](https://pan.baidu.com/s/1c1TtKcw) [[Google drive]](https://drive.google.com/open?id=0B7TOZKXmIjU3ODhaRm8yN2QzRHc)
2017-05-05 13:01:03 +08:00
Using the above IDE models and IDE features, you can reproduce the results with our re-ranking method as follows:
2017-03-22 18:48:01 +08:00
- Market-1501
|Methods |
| -------- | ----- | ---- |
|IDE_ResNet_50 + Euclidean | 78.92% | 55.03%|
2017-05-05 13:01:03 +08:00
|IDE_ResNet_50 + Euclidean + re-ranking | 81.44% | 70.39%|
2017-03-22 18:48:01 +08:00
|IDE_ResNet_50 + XQDA | 77.58% | 56.06%|
2017-05-05 13:01:03 +08:00
|IDE_ResNet_50 + XQDA + re-ranking | 80.70% | 69.98%|
2017-03-22 18:48:01 +08:00
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|
2017-05-05 16:15:28 +08:00
|BOW + XQDA [1] | 7.93% | 7.29%|6.36% | 6.39%|
|BOW + XQDA + re-ranking | 8.93% | 9.94%|8.29% | 8.81%|
2017-05-05 15:12:55 +08:00
|LOMO + XQDA [3] | 14.8% | 13.6%|12.8% | 11.5%|
2017-05-05 13:01:03 +08:00
|LOMO + XQDA + re-ranking | 19.1% | 20.8%|16.6% | 17.8%|
2017-03-22 18:48:01 +08:00
|IDE_CaffeNet + Euclidean | 15.6% | 14.9%| 15.1% | 14.2%|
2017-05-05 13:01:03 +08:00
|IDE_CaffeNet + Euclidean + re-ranking | 19.1% | 21.3%|19.3% | 20.6%|
2017-03-22 18:48:01 +08:00
|IDE_CaffeNet + XQDA | 21.9% | 20.0%|21.1% | 19.0%|
2017-05-05 13:01:03 +08:00
|IDE_CaffeNet + XQDA + re-ranking | 25.9% | 27.8%|26.4% | 26.9%|
2017-03-22 18:48:01 +08:00
|IDE_ResNet_50 + Euclidean | 22.2% | 21.0%|21.3% | 19.7%|
2017-05-05 13:01:03 +08:00
|IDE_ResNet_50 + Euclidean + re-ranking | 26.6% | 28.9%|24.9% | 27.3%|
2017-03-22 18:48:01 +08:00
|IDE_ResNet_50 + XQDA | 32.0% | 29.6%|31.1% | 28.2%|
2017-05-05 13:01:03 +08:00
|IDE_ResNet_50 + XQDA + re-ranking | 38.1% | 40.3%|34.7% | 37.4%|
2017-03-22 18:48:01 +08:00
2017-05-05 13:06:18 +08:00
### References
2017-05-05 15:12:55 +08:00
[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.
2017-05-05 13:14:31 +08:00
2017-05-05 13:06:18 +08:00
[2] Sparse contextual activation for efficient visual re-ranking. Bai, Song and Bai, Xiang. IEEE Transactions on Image Processing. 2016
2017-05-05 15:12:55 +08:00
[3] Person re-identification by local maximal occurrence representation and metric learning. Liao S, Hu Y, Zhu X, et al. In CVPR. 2015
2017-03-22 18:48:01 +08:00
### Contact us
If you have any questions about this code, please do not hesitate to contact us.
[Zhun Zhong ](http://zhunzhong.site )
[Liang Zheng ](http://liangzheng.com.cn )
