Szczegóły publikacji
Opis bibliograficzny
Transfer of learning processes and models in the detection process of animals in crowded herds / Kazimierz Choroś, Farid Kazimov // W: Advances in Computational Collective Intelligence : 18th International Conference, ICCCI 2026 : Heraklion, Crete, Greece, September 23–25, 2026 : proceedings , Pt. 1 / eds. Ngoc Thanh Nguyen, [et al.]. — Cham : Springer Nature, cop. 2027. — ( Communications in Computer and Information Science ; ISSN 1865-0929 ; vol. 3043 ). — ISBN: 978-3-032-37932-0; e-ISBN: 978-3-032-37933-7. — S. 179–195. — Bibliogr., Abstr. — Publikacja dostępna online od: 2026-09-18
Autorzy (2)
- Choroś Kazimierz
- Kazimov Farid
Słowa kluczowe
Dane bibliometryczne
| ID BaDAP | 170458 |
|---|---|
| Data dodania do BaDAP | 2026-10-12 |
| DOI | 10.1007/978-3-032-37933-7_12 |
| Rok publikacji | 2027 |
| Typ publikacji | materiały konferencyjne (aut.) |
| Otwarty dostęp |  |
| Wydawca | Springer |
| Konferencja | International Conference on Computational Collective Intelligence: Semantic Web, Social Networks and Multiagent Systems 2026 |
| Czasopismo/seria | Communications in Computer and Information Science |
Abstract
The detection of animals as well as the counting of herds of crowded animals is still a great challenge in computer vision because of the occlusion, varying animal sizes, background complexity, and the diverse morphology of animal species. Normally, we use a specific model defined exclusively for a given species. However, it requires the gathering of an individual training set for every species. It becomes an almost unrealizable task because of the extremely large number of animal species in nature. The study investigates the opportunity to use transfer learning strategies by adapting models trained on one object class, e.g., people to animals, as well as one animal species into another. Several transfer scenarios are evaluated, and the results indicate scenario-dependent transferability and practical applicability under limited target-data conditions. It was observed that transfer effectiveness is influenced by measurable visual and spatial factors, including object shape, relative scale, and crowding patterns, although the objects do not have to belong to the same species. Unlike prior YOLO-based transfer learning studies that remained within a single domain, our experiments extend transfer learning across species boundaries, offering new insights into how measurable visual and spatial factors influence cross-species detector transferability.
