heynenei / pven

Parsing based vehicle ReID

PVEN

This is the official implementation of article "Parsing-based viewaware embedding network for vehicle ReID"[arxiv], which has been accpeted by CVPR20 as a poster article.

Attention!

Since no test criteria are given in the VERIWild paper, the performance of veriwild provided in the PVEN article does not have the junk image removed. The results with the junk image removed are provided to facilitate comparative experiments by subsequent researchers.

small mAP	small cmc@1	small cmc@5	medium mAP	medium cmc@1	medium cmc@5	large mAP	large cmc@1	large cmc@5
79.80	94.01	98.06	73.91	92.03	97.15	66.20	88.62	95.31

Although the performance is slightly lower after removing the junk image, it is still higher than all comparative methods at the time of this paper, and therefore does not affect the core conclusions.

Requirements

1. python 3.6+
2. torch 1.3.1+

Install

git clone https://github.com/silverbulletmdc/PVEN
cd PVEN
pip install -r requirements.txt
python setup.py install

If you want to modify the code of this project, use the following commands instead

cd PVEN
pip install -r requirements.txt
python setup.py develop

Preparing dataset

Before the pipeline, you should prepare your vehicle ReID dataset first. For each dataset, you need to generate a description pickle file for it, which is a pickled dict with following structure:

{
    "train":[
        {
            "filename": "0001_c001_00016450_0.jpg",
            "image_path": "/data/datasets/VeRi/VeRi/image_train/0001_c001_00016450_0.jpg",
            "id": "0001",
            "cam": "001",
        },
        ...

    ],
    "gallery":[
        ...
    ],
    "query":[
        ...
    ]
}

For different dataset, we have already provided the generating scripts to help you generate the pickle file.

cd examples/preprocess_data
# For VeRi776
python generate_pkl.py veri776 --input-path <VeRi_PATH> --output-path ../outputs/veri776.pkl
# For VERIWild
python generate_pkl.py veriwild --input-path <VeRi_PATH> --output-path ../outputs/veriwild.pkl
# For VehicleID 
python generate_pkl.py vehicleid --input-path <VeRi_PATH> --output-path ../outputs/vehicleid.pkl

Training the parsing model

Convert polygons to parsing masks

As is described in the article, we annotated the parsing information of 3165 images from VeRi776. We just annotate the vertexs of the polygons as the vehicles are composed by several polygons. The details of polygons are in examples/parsing/poly.json. Run following command to convert the polygons to parsing masks

cd examples/parsing
python veri776_poly2mask.py --json-path poly.json --output-path ../outputs/veri776_parsing3165

The parsing masks will be generated in ../outputs/veri776_parsing3165 folder.

Train parsing model

Run following command to train the parsing model

cd examples/parsing
python train_parsing.py --train-set trainval --masks-path ../outputs/veri776_parsing3165 --image-path <VeRi_PATH>/image_train

where the <VeRi_PATH> is the path of your VeRi776 dataset.

Generate parsing masks for ReID dataset

Running the following command to generate masks for the whole ReID dataset and write the mask_path to the dataset pickle file.

cd examples/parsing
python generate_masks.py --model-path best_model_trainval.pth --reid-pkl-path ../outputs/veri776.pkl --output-path ../outputs/veri776_masks

where the <PKL_PATH> is the generated pickle file above.

Train PVEN

Run the following model to train PVEN.

cd examples/parsing_reid
# For VeRi776
CUDA_VISIBLE_DEVICES=0 python main.py train -c configs/veri776_b64_pven.yml 
# For vehicleid, use 8 GPUs to train
python main.py train -c configs/vehicleid_b256_pven.yml 
# For VERIWild, use 8 GPUs to train
python main.py train -c configs/veriwild_b256_224_pven.yml 

Pretrained Models

We provide the pretrained parsing model, VeRi776 ReID model and VERIWild ReID model ( the classification layer has been removed ) for your convinient. You can download it from the following link: Link: https://pan.baidu.com/s/1Q2NMVfGZPCskh-E6vmy9Cw password: iiw1

Evaluate PVEN

cd examples/parsing_reid
# For VeRi776
python main.py eval -c configs/veri776_b64_parsing.yml

# For VERIWild
## small
python main.py eval -c configs/veriwild_b256_224_pven.yml
## medium
python main.py eval -c configs/veriwild_b256_224_pven.yml test.ext _5000
## Large
python main.py eval -c configs/veriwild_b256_224_pven.yml test.ext _10000

Citation

If you found our method helpful in your research, please cite our work in your publication.

@inproceedings{meng2020parsing,
  title={Parsing-based View-aware Embedding Network for Vehicle Re-Identification},
  author={Meng, Dechao and Li, Liang and Liu, Xuejing and Li, Yadong and Yang, Shijie and Zha, Zheng-Jun and Gao, Xingyu and Wang, Shuhui and Huang, Qingming},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={7103--7112},
  year={2020}
}