Implementation of the cityAI CVPR 2020 paper
Our Code is based on the bags of tricks repo.
- Description
- Orientation dataset generation
- Dataset download
- Training the models from the paper
- Download the models from the paper
- Generate submission and files for visualisation
- Citation
To generate the orientation dataset download the vehicleX code
Move the script unity3D/GenerateOrientationDataWithDistractors.cs in the project and change the paths to the background images you want to use.
Click hold and move the script on the camera. Delete all opject including the lights in the scene. Run.
After generating the data use the vehicleX SPGAN to domain adapt the orientation data.
Or you can download the dataset.
Download the cityAI datasets and the synthetic orient data from link and place it in data/data/cityAI.
The labels for the orientation data is here.
data
--- data
--- --- cityAI
--- --- --- real
--- --- --- synthetic
--- --- orient
--- --- --- synthetic
--- --- --- real
To train the I2I 256x256 run
2020-aicitychallenge-IOSB-VeRi/logs$ WANDB_MODE=dryrun WANDB_CONFIG_PATHS=../config/config_singlenet_256x256.yaml CUDA_VISIBLE_DEVICES=0 taskset -c 6-10 python3 /net/merkur/storage/deeplearning/users/eckvik/WorkspaceNeu/VReID/2020-aicitychallenge-IOSB-VeRi/tools/train.py
Similar with the I2I 320x320 model and with the V2V models.
2020-aicitychallenge-IOSB-VeRi/logs$ WANDB_MODE=dryrun WANDB_CONFIG_PATHS=../config/resnet152_i2i_320x320.yaml CUDA_VISIBLE_DEVICES=0 taskset -c 6-10 python3 /net/merkur/storage/deeplearning/users/eckvik/WorkspaceNeu/VReID/2020-aicitychallenge-IOSB-VeRi/tools/train.py
2020-aicitychallenge-IOSB-VeRi/logs$ WANDB_MODE=dryrun WANDB_CONFIG_PATHS=../config/resnet152_v2v_256x256.yaml CUDA_VISIBLE_DEVICES=0 taskset -c 6-10 python3 /net/merkur/storage/deeplearning/users/eckvik/WorkspaceNeu/VReID/2020-aicitychallenge-IOSB-VeRi/tools/train.py
2020-aicitychallenge-IOSB-VeRi/logs$ WANDB_MODE=dryrun WANDB_CONFIG_PATHS=../config/resnet152_v2v_320x320.yaml CUDA_VISIBLE_DEVICES=0 taskset -c 6-10 python3 /net/merkur/storage/deeplearning/users/eckvik/WorkspaceNeu/VReID/2020-aicitychallenge-IOSB-VeRi/tools/train.py
To train the orientation model
2020-aicitychallenge-IOSB-VeRi/logs$ WANDB_MODE=dryrun WANDB_CONFIG_PATHS=../config/orient_net.yaml CUDA_VISIBLE_DEVICES=0 taskset -c 6-10 python3 /net/merkur/storage/deeplearning/users/eckvik/WorkspaceNeu/VReID/2020-aicitychallenge-IOSB-VeRi/tools/train_regression.py
the results will be saved in logs/wandb/
Or you can download the pretrained models.
Download the pretrained models link and move them to the directory pretrained.
To generate the submit.txt and the files for visualisation run
Before ensembling
CUDA_VISIBLE_DEVICES=3 WANDB_CONFIG_PATHS=2020-aicitychallenge-IOSB-VeRi/config/config_singlenet_256x256.yaml WANDB_MODE=dryrun python3 2020-aicitychallenge-IOSB-VeRi/tools/generate_txt_files_for_visualisation.py -wp
/net/merkur/storage/deeplearning/users/eckvik/WorkspaceNeu/VReID/VehicleReID/logs/cityai/2020/wandb/run-20200325_181358-7zlgl6yc_my_logs_justfold1/resnet152_model_63.pth
-rp
/net/merkur/storage/deeplearning/users/eckvik/WorkspaceNeu/VReID/2020-aicitychallenge-IOSB-VeRi/logs/wandb/dryrun-20200507_112944-ctijgkkg_my_logs_justfold1
After ensembling
CUDA_VISIBLE_DEVICES=3 WANDB_CONFIG_PATHS=2020-aicitychallenge-IOSB-VeRi/config/config_singlenet_256x256.yaml WANDB_MODE=dryrun python3 2020-aicitychallenge-IOSB-VeRi/tools/ensembling_different_models.py
ensembling_different_models.py generates the submit file
You need an input folder with images and an empty output folder. Download the weights, select a config file and run:
CUDA_VISIBLE_DEVICES=3 .../tools/generate_feats.py --input ../input --output ../output --config ../config/resnet152_i2i_256x256.yaml --weights ..../resnet152_model_63.pth
@InProceedings{Eckstein_2020_CVPR_Workshops,
author = {Eckstein, Viktor and Schumann, Arne and Specker, Andreas},
title = {Large Scale Vehicle Re-Identification by Knowledge Transfer From Simulated Data and Temporal Attention},
booktitle = {The IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
month = {June},
year = {2020}
}
Bags of tricks
@InProceedings{Luo_2019_CVPR_Workshops,
author = {Luo, Hao and Gu, Youzhi and Liao, Xingyu and Lai, Shenqi and Jiang, Wei},
title = {Bag of Tricks and a Strong Baseline for Deep Person Re-Identification},
booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
month = {June},
year = {2019}
}
@ARTICLE{Luo_2019_Strong_TMM,
author={H. {Luo} and W. {Jiang} and Y. {Gu} and F. {Liu} and X. {Liao} and S. {Lai} and J. {Gu}},
journal={IEEE Transactions on Multimedia},
title={A Strong Baseline and Batch Normalization Neck for Deep Person Re-identification},
year={2019},
pages={1-1},
doi={10.1109/TMM.2019.2958756},
ISSN={1941-0077},
}
VehicleX
@article{yao2019simulating,
title={Simulating Content Consistent Vehicle Datasets with Attribute Descent},
author={Yao, Yue and Zheng, Liang and Yang, Xiaodong and Naphade, Milind and Gedeon, Tom},
journal={arXiv preprint arXiv:1912.08855},
year={2019}
}
@inproceedings{tang2019pamtri,
title={Pamtri: Pose-aware multi-task learning for vehicle re-identification using highly randomized synthetic data},
author={Tang, Zheng and Naphade, Milind and Birchfield, Stan and Tremblay, Jonathan and Hodge, William and Kumar, Ratnesh and Wang, Shuo and Yang, Xiaodong},
booktitle={Proceedings of the IEEE International Conference on Computer Vision},
pages={211--220},
year={2019}
}
@inproceedings{image-image18,
author = {Weijian Deng and
Liang Zheng and
Qixiang Ye and
Guoliang Kang and
Yi Yang and
Jianbin Jiao},
title = {Image-Image Domain Adaptation with Preserved Self-Similarity and Domain-Dissimilarity
for Person Re-identification},
booktitle = {CVPR},
year = {2018},
}