Efficient Multi-Class Object Re-Identification Approach with A Unified Model

Re-identification (ReID) is a critical field in com-puter vision. In previous studies, ReID models have primarily focused on identifying specific objects, such as persons or vehi-cles. However, handling diverse object types poses a significant challenge in real-world applications. This paper proposes an advanced multi-task
$ReID$
model that identifies and classifies multiple object types within a single architecture. The proposed method allows our model to achieve high performance and gener-alizability across various object categories, including humans and vehicles. Our proposed model combines the strengths of multiple loss functions and incorporates several supporting techniques. We evaluate our model using two popular datasets, Market-l50l and VeRi-776. Our proposed model achieves 97.8% Rank@1 on the VeRi-776 dataset and 96.7% Rank@1 on the Market-l50l dataset in single-task training and evaluation. With our multi-class object re-identification approach, we simultaneously achieved an accuracy of 100% for the classification problem, 95.3% Rank@1 using the evaluation without classification information method, 95.4 % Rank @ 1 using the evaluation based on the classification method on the VeRi-776 dataset, and 9S.2% Rank@1 on the Market-l50l dataset. Code is available at this https URL.