In this package, we provide our training and testing code written in pytorch for the paper A Discriminatively Learned CNN Embedding for Person Re-identification.
Compared with the original version, I do some modification:
- I use
x*y
instead of(x-y)^2
asSquare Layer
. (We do not need to worry about the scale ofx
andy
.) - I add the bottle-neck fully-connected layer for classification. I use the
512-dim
fully-connected feature as pedestrian descriptor. - I tune some hyperparameters.
- On MSMT-17, we arrive Rank@1:0.604769 Rank@5:0.761815 Rank@10:0.815593 mAP:0.315827.
We arrived Rank@1=88.66%, mAP=72.58% with ResNet-50. The code is largely borrowed from my another repo strong Pytorch baseline . Here we provide hyperparameters and architectures, that were used to generate the result. Some of them (i.e. learning rate) are far from optimal. Do not hesitate to change them and see the effect.
Any suggestion is welcomed.
This code is ONLY released for academic use.
- Zhedong Zheng The original Matconvnet version in the paper. (
)
- Weihang Chen also realizes our paper in Keras. (
)
- Xuanyi Dong also realizes our paper in Caffe. (
)
- Zhun Zhong provides a extensive Caffe baseline code. You may check it. (
)
- Zhedong Zheng provides a strong Pytorch baseline (
)
You may learn more from model.py
.
- Python 3.6
- GPU Memory >= 6G
- Numpy
- Pytorch 0.3+
(Some reports found that updating numpy can arrive the right accuracy. If you only get 50~80 Top1 Accuracy, just try it.) We have successfully run the code based on numpy 1.12.1 and 1.13.1 .
- Install Pytorch from http://pytorch.org/
- Install Torchvision from the source
git clone https://github.com/pytorch/vision
cd vision
python setup.py install
Because pytorch and torchvision are ongoing projects.
Here we noted that our code is tested based on Pytorch 0.3.0/0.4.0 and Torchvision 0.2.0.
Download Market1501 Dataset
Preparation: Put the images with the same id in one folder. You may use
python prepare.py
Remember to change the dataset path to your own path.
Futhermore, you also can test our code on DukeMTMC-reID Dataset. Our baseline code is not such high on DukeMTMC-reID Rank@1=64.23%, mAP=43.92%. Hyperparameters are need to be tuned.
To save trained model, we make a dir.
mkdir model
Train a model by
python train_new.py --gpu_ids 0 --name ft_ResNet50 --alpha 1.0 --batchsize 32 --data_dir your_data_path
--gpu_ids
which gpu to run.
--name
the name of model.
--data_dir
the path of the training data.
--batchsize
batch size.
--erasing_p
random erasing probability.
--alpha
the weight of the verification loss.
Train a model with random erasing by
python train_new.py --gpu_ids 0 --name ft_ResNet50 --train_all --batchsize 32 --data_dir your_data_path --erasing_p 0.5
Use trained model to extract feature by
python test.py --gpu_ids 0 --name ft_ResNet50 --test_dir your_data_path --which_epoch 59
--gpu_ids
which gpu to run.
--name
the dir name of trained model.
--which_epoch
select the i-th model.
--data_dir
the path of the testing data.
python evaluate.py
It will output Rank@1, Rank@5, Rank@10 and mAP results.
You may also try evaluate_gpu.py
to conduct a faster evaluation with GPU.
For mAP calculation, you also can refer to the C++ code for Oxford Building. We use the triangle mAP calculation (consistent with the Market1501 original code).
Please cite this paper in your publications if it helps your research:
@article{zheng2016discriminatively,
title={A Discriminatively Learned CNN Embedding for Person Re-identification},
author={Zheng, Zhedong and Zheng, Liang and Yang, Yi},
doi={10.1145/3159171},
journal={ACM Transactions on Multimedia Computing Communications and Applications},
year={2017}
}