WildSR-DETR: Super-Resolution-Aware End-to-End Object Detection for Wildlife in the Wild
Accurate wildlife detection from camera trap imagery remains challenging due to frequent low-resolution inputs, dense environments, and scale variation. This study introduces WildSR-DETR, a novel architecture that integrates RT-DETR with a super-resolution reconstruction module, enabling object detection directly on enhanced wildlife images. Experiments on the NTLNP dataset demonstrate that our model achieves a precision of 98.2%, recall of 98.5%, and mAP@50 of 98.9%, outperforming state-of-the-art models with competitive inference speed. This framework provides a practical foundation for large-scale wildlife monitoring and ecological conservation.
Authors
- Jianghan Tao (ORCID: https://orcid.org/0000-0002-0892-836X)
- Xue Feng (ORCID: https://orcid.org/0000-0002-9057-1549)
- Boheng Li
- Yong Sun (ORCID: https://orcid.org/0000-0001-5246-0971)
- Zhijie Zhang (ORCID: https://orcid.org/0000-0002-0139-3892)
- Xiaoyu Zhang (ORCID: https://orcid.org/0000-0001-8905-0562)
- Chufeng Sun
- Danhua Liu (ORCID: https://orcid.org/0000-0001-6823-6691)
- Jialuo Du
- Xiaoyu Guo (ORCID: https://orcid.org/0009-0008-1773-1789)
Institutions
- Johns Hopkins University (US)
- Waseda University (JP)
- Hosei University (JP)
- China University of Petroleum, Beijing (CN)
- Johns Hopkins Medicine (US)
- Johns Hopkins University Applied Physics Laboratory (US)
- Sophia University (JP)
- Beijing City University (CN)
- The University of Tokyo (JP)
Publication Details
- Published
- 2025-06-06
- DOI
- https://doi.org/10.1109/iccgiv65419.2025.11085090
- Primary Topic
- Advanced Image and Video Retrieval Techniques
- Type
- article
- Field-Weighted Citation Impact
- 0.00
Funders
- Japan Science and Technology Agency