Enhancing Long-Term Re-Identification Robustness Using Synthetic Data: A Comparative Analysis

TL;DR

This work addresses long-term re-identification under material aging in an industrial setting by introducing the pallet-block-2696 dataset, a 2,696-image open-source collection captured over four months with natural aging and damage. It benchmarks three gallery-expansion strategies (T00, T01, T02) and evaluates pretrained re-id models (ResNet50, PCB, OSNet) trained on real and synthetic data, using $mAP$ and Rank-$k$ metrics. The study shows that ongoing gallery updates can boost Rank-$1$ accuracy by about $24\%$, and incorporating synthetic data can improve performance by up to $13\%$ relative to real-data training alone, though robustness over time remains a challenge. These findings advance practical re-id in logistics by enabling aging-aware identification and providing an open dataset to accelerate further research in synthetic aging and lifelong gallery maintenance.

Abstract

This contribution explores the impact of synthetic training data usage and the prediction of material wear and aging in the context of re-identification. Different experimental setups and gallery set expanding strategies are tested, analyzing their impact on performance over time for aging re-identification subjects. Using a continuously updating gallery, we were able to increase our mean Rank-1 accuracy by 24%, as material aging was taken into account step by step. In addition, using models trained with 10% artificial training data, Rank-1 accuracy could be increased by up to 13%, in comparison to a model trained on only real-world data, significantly boosting generalized performance on hold-out data. Finally, this work introduces a novel, open-source re-identification dataset, pallet-block-2696. This dataset contains 2,696 images of Euro pallets, taken over a period of 4 months. During this time, natural aging processes occurred and some of the pallets were damaged during their usage. These wear and tear processes significantly changed the appearance of the pallets, providing a dataset that can be used to generate synthetically aged pallets or other wooden materials.

Figures (3)

• Figure 1: Example images of ID 26. The series in each row was recorded on the following days (from top to down): recording day 1, recording day 4 and recording day 14a. Notably, the series from day 14a shows the damaged pallet blocks.
• Figure 2: The full aging process of ID 0 over four months.
• Figure 3: The aging process of ID 2, the upper row showing a pronounced branding pattern decay over four months, and the lower row showing a similar aging process, generated synthetically (taken from reidrobustness2023).