- author = "Yiming Li and Lin Shang",
- title = "{Re-ID BUFF}: An Enhanced Similarity Measurement Based on Genetic Programming for Person Re-identification",
- booktitle = "Proceedings of the 2021 Genetic and Evolutionary Computation Conference Companion",
- year = "2021",
- editor = "Francisco Chicano and Alberto Tonda and Krzysztof Krawiec and Marde Helbig and Christopher W. Cleghorn and Dennis G. Wilson and Georgios Yannakakis and Luis Paquete and Gabriela Ochoa and Jaume Bacardit and Christian Gagne and Sanaz Mostaghim and Laetitia Jourdan and Oliver Schuetze and Petr Posik and Carlos Segura and Renato Tinos and Carlos Cotta and Malcolm Heywood and Mengjie Zhang and Leonardo Trujillo and Risto Miikkulainen and Bing Xue and Aneta Neumann and Richard Allmendinger and Fuyuki Ishikawa and Inmaculada Medina-Bulo and Frank Neumann and Andrew M. Sutton",
- pages = "255--256",
- address = "internet",
- series = "GECCO '21",
- month = jul # " 10-14",
- organisation = "SIGEVO",
- publisher = "Association for Computing Machinery",
- publisher_address = "New York, NY, USA",
- keywords = "genetic algorithms, genetic programming, Person reidentification,: Poster",
- isbn13 = "978-1-4503-8351-6",
-
DOI = "
doi:10.1145/3449726.3459432",
- size = "2 pages",
- abstract = "Person re-identification (re-ID) is a fundamental link to ensure successful cross-camera tracking. Its overall process is divided into two stages, feature extraction and matching. In the matching stage,the simple Euclid measurement has limitations in score due to the lack of consideration of the vector directions. The majority of recent contributions focus on the use of neighborhood information of the query image, which makes the calculation process more complex. In addition, these similarity measurements are predefined and therefore evaluated. In this paper, we propose a re-ID similarity measurement based on Genetic Programming (GP). The similarity measurement formula is automatically evolved and constructed through the optimization using specific function set, terminal set and fitness function targeting re-ID. For the training process of GP,we propose the feature triplet-dataset like the triplet-loss based on the existing re-ID datasets. We conduct sufficient experiments on three benchmark datasets, comparing with features of different quality and various measurements. The proposed method has a better effect than the Euclid measurement and achieves a general improvement on the mAP and rank-1 of combination with other metrics. Therefore, our method can be used as a gain buff to enhance the score of other methods on the original basis.",
-
notes = "Nanjing University
GECCO-2021 A Recombination of the 30th International Conference on Genetic Algorithms (ICGA) and the 26th Annual Genetic Programming Conference (GP)",
