Learnable Privacy-Preserving Anonymization for Pedestrian Images PDF Junwu Zhang, Mang Ye, Yao Yang ACM MM, 2022
- A reversible visual anonymization framework for Re-ID on hybrid images composed of raw and protected images.
- A progressive training strategy, namely supervision upgradation, which improves Re-ID performance while retaining enough unrecognizability.
Pretrained models will be published later.
The detailed results can be found in the main paper and the results of widely used non-protective Re-ID model AGW can be found here.
| Blur | Mosaic | Noise | |
|---|---|---|---|
| Base | 84.8/40.1/18.3/67.3 | 87.4/75.3/70.9/64.3 | 75.9/50.8/41.7/68.7 |
| Ours w/o U | 83.1/68.3/46.8/75.8 | 86.3/80.1/75.1/73.2 | 90.4/77.5/67.5/84.4 |
| Ours | 91.6/88.2/82.5/89.2 | 89.4/88.5/86.8/87.0 | 91.7/83.5/83.8/91.2 |
The four values separated by '/' are corresponding to the rank-1 accuracy under raw/crossed/protected settings in the main paper.
| Blur | Mosaic | Noise | |
|---|---|---|---|
| Base | 30.5/21.3/16.2/15.2 | 25.0/16.3/14.6/10.6 | 24.1/20.1/18.5/18.2 |
| Ours w/o U | 43.6/28.4/10.0/14.7 | 34.3/26.2/24.6/20.5 | 41.9/36.0/29.4/31.3 |
| Ours | 51.5/51.1/50.5/48.7 | 48.6/49.1/48.5/48.1 | 46.9/48.1/46.2/46.0 |
| Blur | Mosaic | Noise | |
|---|---|---|---|
| Base | 30.4/14.6/10.4/8.2 | 28.5/17.7/15.1/8.8 | 14.0/9.1/8.6/8.1 |
| Ours w/o U | 31.6/13.8/8.9/14.3 | 24.9/24.2/19.6/12.1 | 28.1/30.4/30.4/31.9 |
| Ours | 41.9/39.2/35.3/33.2 | 38.8/37.8/30.4/27.6 | 35.8/36.4/37.9/41.9 |
Install
Clone repo and install requirements.txt in a Python>=3.7.10 environment, including PyTorch>=1.8.
git clone https://github.com/whuzjw/privacy-reid.git
cd privacy-reid
pip install -r requirements.txtPrepare Datasets
- Download Market1501, MSMT17 and CUHK03 datasets from https://www.liangzheng.org/Project/project_reid.html to <path_to_root_of_datasets>.
- Extract dataset and rename to
market1501. The data structure would like:
<path_to_root_of_datasets/market1501>
market1501
bounding_box_test/
bounding_box_train/
query/
- Split Market1501 training set into a smaller training set (80%) and a validation set (20%). Furthermore, the validation set is split into a validation gallery set (80%) and a validation query set (20%).
python3 data/train_split.py --dataset_root_dir '<path_to_root_of_datasets>' --dataset_name 'market1501'MSMT17 can be split by changing 'market1501' to 'msmt17', CUHK03 can be split using Matlab.
The data structure would like:
<path_to_root_of_datasets/market1501_val>
market1501
bounding_box_test/
bounding_box_train/
query/
bounding_box_val_gallery/
bounding_box_val_query/
Baseline
python3 tools/main.py --config_file='configs/AGW_baseline.yml' DATASETS.ROOT_DIR "<path_to_root_of_datasets>" DATASETS.NAMES "('market1501')" INPUT.TYPE "mosaic" INPUT.RADIUS "24.0" OUTPUT_DIR "('./log/market1501/market1501-base-mosaic')" MODEL.MODE "C"Replace <path_to_root_of_datasets> by your path to the root of datasets, e.g., /data/Dataset.
For other datasets, replace 'market1501' by 'msmt17' or 'cuhk03'.
For other desensitization methods, INPUT.TYPE and INPUT.RADIUS can be changed from mosaic 24.0×24.0 to blur 12.0×12.0 or noise 0.5.
Our Model w/o Supervision Upgradation
python3 tools/main.py --config_file='configs/AGW_baseline.yml' DATASETS.ROOT_DIR "<path_to_root_of_datasets>" DATASETS.NAMES "('market1501')" INPUT.TYPE "mosaic" INPUT.RADIUS "24.0" OUTPUT_DIR "('./log/market1501/market1501-wosu-mosaic')"Our Full Model
python3 tools/main.py --config_file='configs/AGW_baseline.yml' DATASETS.ROOT_DIR "<path_to_root_of_datasets>" DATASETS.NAMES "('market1501_val')" INPUT.TYPE "mosaic" INPUT.RADIUS "24.0" OUTPUT_DIR "('./log/market1501/market1501-full-mosaic')" MODEL.VAL_R1 "<R1_wosu-mosaic>"Replace <R1_mosaic-wo> by the rank-1 accuracy of a trained market1501-wosu-mosaic model under the setting of raw images as query and mosaic images as gallery.
The testing command is just the incremental version of the training command, which is prefixed with the training command and suffixed with TEST.EVALUATE_ONLY, MODEL.PRETRAIN_CHOICE, and MODEL.PRETRAIN_DIR.
Baseline
python3 tools/main.py --config_file='configs/AGW_baseline.yml' DATASETS.ROOT_DIR "<path_to_root_of_datasets>" DATASETS.NAMES "('market1501')" INPUT.TYPE "mosaic" INPUT.RADIUS "24.0" OUTPUT_DIR "('./log/market1501/test/market1501-base-mosaic')" MODEL.MODE "C" TEST.EVALUATE_ONLY "('on')" MODEL.PRETRAIN_CHOICE "('self')" MODEL.PRETRAIN_DIR "./log/market1501/market1501-base-mosaic"Our Model w/o Supervision Upgradation
python3 tools/main.py --config_file='configs/AGW_baseline.yml' DATASETS.ROOT_DIR "<path_to_root_of_datasets>" DATASETS.NAMES "('market1501')" INPUT.TYPE "mosaic" INPUT.RADIUS "24.0" OUTPUT_DIR "('./log/market1501/test/market1501-wosu-mosaic')" TEST.EVALUATE_ONLY "('on')" MODEL.PRETRAIN_CHOICE "('self')" MODEL.PRETRAIN_DIR "./log/market1501/market1501-wosu-mosaic"Our Model w/o Supervision Upgradation
python3 tools/main.py --config_file='configs/AGW_baseline.yml' DATASETS.ROOT_DIR "<path_to_root_of_datasets>" DATASETS.NAMES "('market1501')" INPUT.TYPE "mosaic" INPUT.RADIUS "24.0" OUTPUT_DIR "('./log/market1501/test/market1501-full-mosaic')" TEST.EVALUATE_ONLY "('on')" MODEL.PRETRAIN_CHOICE "('self')" MODEL.PRETRAIN_DIR "./log/market1501/market1501-full-mosaic"Please kindly cite this paper in your publications if it helps your research:
@article{acmmm21reidprivacy,
title={Learnable Privacy-Preserving Anonymization for Pedestrian Images},
author={Junwu, Zhang and Mang, Ye and Yao, Yang},
journal={ACM MM},
year={2022},
}
Contact: [email protected]