Towards Cross-Modal Preservation, Consistency and Alignment for Privacy-Preserving Visible-Infrared Person Re-Identification

Abstract

Privacy-preserving Person Re-Identification (PP-ReID) addresses the core privacy-utility trade-off in Re-ID by retrieving a person across multiple non-overlapping cameras while applying anonymization techniques to protect sensitive information. However, prior PP-ReID studies are confined to single-modality visible scenarios, as 24-hour surveillance systems require robust cross-modal visibleinfrared (VI) capabilities. Extending PP-ReID to the crossmodal VI setting is therefore crucial. Accordingly, we introduce a new task: Privacy-Preserving Visible-Infrared Person Re-Identification (PP-VI-ReID). This task presents two severe challenges: 1) Crude anonymization strategies destroy identity-critical information and disrupt cross-modal alignment 2) The anonymization process creates inconsistent distortions across modalities. It disrupts color-based textures in visible images while obscuring thermal contours in infrared images. This inconsistency with modality gap forms a Mixed Gap. To overcome these challenges, we propose a framework, the Precise Privacy-preserving and Alignment Network (PPA) with two components: 1) A Keypoint-Preserving Regularization (KPR) module leverages human pose as a prior to guide structure-aware anonymization, preserving essential body features. 2) A Differential Consistency-guided Modality Alignment (DCMA) treats anonymization perturbations not as varying noise, but as a stable, learnable offset, facilitating robust alignment between raw and anonymized features across modalities. Experiments on SYSU-MM01 and RegDB validate our framework, establishing a strong baseline.

Framework

---

Environment

conda create -n ppa python=3.8
conda activate ppa
pip install -r requirements.txt

Datasets

• (1) RegDB Dataset [3]: The RegDB dataset can be downloaded from this website by submitting a copyright form.

  • (Named: "Dongguk Body-based Person Recognition Database (DBPerson-Recog-DB1)" on their website).

• (2) SYSU-MM01 Dataset [4]: The SYSU-MM01 dataset can be downloaded from this website.

  • run python pre_process_sysu.py to pepare the dataset, the training data will be stored in ".npy" format.

Teacher Checkpoint

This project also uses a teacher model. Please define cfg.PRETRAINED_TEACHER in config/configs.py as the folder that stores teacher checkpoints:

cfg.PRETRAINED_TEACHER = './pretrained'

Prepare teacher parameter files in this folder:

pretrained/
├── teacher_model_sysu.t
└── teacher_model_regdb.t

The code automatically selects the teacher checkpoint by dataset name:

• DATASETS.NAMES == 'sysu' -> teacher_model_sysu.t
• DATASETS.NAMES == 'regdb' -> teacher_model_regdb.t

You can download the teacher checkpoints from Google Drive. After downloading, place the checkpoint files in the directory specified by cfg.PRETRAINED_TEACHER.

train

python main.py MODEL.DEVICE "cuda:7" MODEL.DEVICE_ID "('7')" DATASETS.ROOT_DIR "/data0/yb_data/sysu/" DATASETS.NAMES "('sysu')" INPUT.TYPE "mosaic" INPUT.RADIUS "24.0" OUTPUT_DIR "('/data1/yb_log/csreid/sysu/sysu-mosaic-full')" MODEL.VAL_R1 "0.15" MODEL.DCMA "on" MODEL.DCMA_LAMBDA "0.1" MODEL.POSE_GUIDE "on" MODEL.POSE_LAMBDA "1.0"

python main.py MODEL.DEVICE "cuda:4" MODEL.DEVICE_ID "('4')" DATASETS.ROOT_DIR "/data0/yb_data/sysu/" DATASETS.NAMES "('sysu')" INPUT.TYPE "noise" INPUT.RADIUS "0.5" OUTPUT_DIR "('/data1/yb_log/csreid/sysu/sysu-noise-full')" MODEL.VAL_R1 "0.15" MODEL.DCMA "on" MODEL.DCMA_LAMBDA "0.1" MODEL.POSE_GUIDE "on" MODEL.POSE_LAMBDA "1.0"

python main.py MODEL.DEVICE "cuda:7" MODEL.DEVICE_ID "('7')" DATASETS.ROOT_DIR "/data0/yb_data/sysu/" DATASETS.NAMES "('sysu')" INPUT.TYPE "blur" INPUT.RADIUS "12.0" OUTPUT_DIR "('/data1/yb_log/csreid/sysu/sysu-blur-full')" MODEL.VAL_R1 "0.15" MODEL.DCMA "on" MODEL.DCMA_LAMBDA "0.1" MODEL.POSE_GUIDE "on" MODEL.POSE_LAMBDA "1.0"

Test

python test.py \
  INPUT.TYPE "mosaic" \
  INPUT.RADIUS "24.0" \
  OUTPUT_DIR "./log/sysu/test/sysu-mosaic-test" \
  TEST.EVALUATE_ONLY "('on')" \
  MODEL.PRETRAIN_CHOICE "('self')" \
  MODEL.PRETRAIN_DIR "./log/sysu/sysu-mosaic"

python testreg.py \
  DATASETS.ROOT_DIR "data/regdb/" \
  DATASETS.NAMES "('regdb')" \
  INPUT.TYPE "mosaic" \
  INPUT.RADIUS "24.0" \
  OUTPUT_DIR "./log/regdb/test/regdb-mosaic-test" \
  TEST.EVALUATE_ONLY "('on')" \
  MODEL.PRETRAIN_CHOICE "('self')" \
  MODEL.PRETRAIN_DIR "./log/regdb/regdb-mosaic"

output

OUTPUT_DIR/
├── events.out.tfevents.xxx  # TensorBoard
├── epoch_best_net_C.pth     # reid_network
├── epoch_best_net_G.pth     # generator
├── epoch_best_net_D1.pth    
└── log.txt                  # log

Contact

x2471722964@outlook.com