DOI QR코드

DOI QR Code

A Synthetic Data Generation and Learning Framework Using Top-down View Transformation for Situation Awareness Enhancement in Unmanned Battlefield Systems

전장 상황 무인 시스템 상황인지 향상을 위한 Top-down 변환 기반 합성 데이터 생성 및 학습 기법

  • Yerin Kim (Electronics and Telecommunications Research Institute (ETRI), Artificial Intelligence Major, University of Science and Technology (UST)) ;
  • Jaeuk Baek (Electronics and Telecommunications Research Institute (ETRI)) ;
  • Sungwoo Jun (Electronics and Telecommunications Research Institute (ETRI)) ;
  • Chang-Eun Lee (Electronics and Telecommunications Research Institute (ETRI), Artificial Intelligence Major, University of Science and Technology (UST))
  • Received : 2026.03.23
  • Accepted : 2026.07.10
  • Published : 2026.08.31

Abstract

Military robots and unmanned systems require training data with diverse viewpoints for robust object detection and situational awareness. Top-down view images are difficult to acquire, resulting in limited viewpoint diversity. In this paper, we propose a pipeline that transforms ground view images into top-down view images while preserving structural consistency. The framework employs FLUX.2 with Low-Rank Adaptation (LoRA) for parameter-efficient viewpoint adaptation. Instead of updating the entire network, we apply LoRA only to the Cross-Attention layers, which are critical for viewpoint conditioning, while freezing the pretrained network. We demonstrate that this approach effectively performs viewpoint transformation in data-limited military domains using only a small-scale training dataset. In addition, we employ conditional generation with ground view inputs, prompts, and target images. Our approach adapts a large pretrained model using only small datasets to generate structurally consistent top-down synthetic images. Structural consistency analysis based on LightGlue correspondences and CLIPPER-based geometric filtering yields a maximum inlier ratio of 52.7%. Finally, training with the synthetic data improves the mean Average Precision (mAP) by 6.5 percentage points over the original-data-only baseline.

Keywords

Acknowledgement

This work was supported by Korea Research Institute for defense Technology planning and advancement (KRIT) grant funded by the Korea government (DAPA (Defense Acquisition Program Administration)) (No. 21-107-E00-009-02, "Realtime complex battlefield situation awareness technology").

References

  1. P. Isola, J.-Y. Zhu, T. Zhou, and A. A. Efros, "Image-to-Image Translation with Conditional Adversarial Networks," in IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Honolulu, HI, USA, pp. 1125-1134, 2017, DOI: 10.1109/CVPR.2017.632.
  2. J.-Y. Zhu, T. Park, P. Isola, and A. A. Efros, "Unpaired Imageto-Image Translation Using Cycle-Consistent Adversarial Networks," in IEEE International Conference on Computer Vision (ICCV), Venice, Italy, pp. 2223-2232, 2017, DOI: 10.1109/ICCV.2017.244.
  3. D. Kothandaraman, T. Zhou, M. Lin, and D. Manocha, "Aerial Diffusion: Text Guided Ground-to-Aerial View Translation from a Single Image using Diffusion Models," in SIGGRAPH Asia 2023 Technical Communications, Sydney, Australia, 2023, DOI:10.1145/3610543.3626177.
  4. A. Arrabi, X. Zhang, W. Sultani, C. Chen, and S. Wshah, "CrossView Meets Diffusion: Aerial Image Synthesis with Geometry and Text Guidance," in 2025 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), Tucson, AZ, USA, 2025, [Online], https://openaccess.thecvf.com/content/WACV2025/html/Arrabi_Cross-View_Meets_Diffusion_Aerial_Image_Synthesis_with_Geometry_and_Text_WACV_2025_paper.html.
  5. J. Ye, J. He, W. Li, Z. Lv, Y. Lin, J. Yu, H. Yang, and C. He, "Leveraging BEV Paradigm for Ground-to-Aerial Image Synthesis," in IEEE/CVF International Conference on Computer Vision (ICCV), Honolulu, HI, USA, pp. 28451-28461, 2025, [Online], https://openaccess.thecvf.com/content/ICCV2025/html/Ye_Leveraging_BEV_Paradigm_for_Ground-to-Aerial_Image_Synthesis_ICCV_2025_paper.html.
  6. B. Mildenhall, P. P. Srinivasan, M. Tancik, J. T. Barron, R. Ramamoorthi, and R. Ng, "NeRF: Representing Scenes as Neural Radiance Fields for View Synthesis," in European Conference on Computer Vision (ECCV), Glasgow, UK, pp. 405-421, 2020, DOI: 10.1007/978-3-030-58452-8_24.
  7. T. Müller, A. Evans, C. Schied, and A. Keller, "Instant Neural Graphics Primitives with a Multiresolution Hash Encoding," ACM Transactions on Graphics, vol. 41, no. 4, Article 102, 2022, DOI: 10.1145/3528223.3530127.
  8. J. T. Barron, B. Mildenhall, D. Verbin, P. P. Srinivasan, and P. Hedman, "Zip-NeRF: Anti-Aliased Grid-Based Neural Radiance Fields," in IEEE/CVF International Conference on Computer Vision (ICCV), Paris, France, pp. 19697-19705, 2023, DOI:10.1109/ICCV51070.2023.01804.
  9. B. Kerbl, G. Kopanas, T. Leimkühler, and G. Drettakis, "3D Gaussian Splatting for Real-Time Radiance Field Rendering," ACM Transactions on Graphics, vol. 42, no. 4, Article 139, 2023, DOI: 10.1145/3592433.
  10. R. Liu, R. Wu, B. Van Hoorick, P. Tokmakov, S. Zakharov, and C. Vondrick, "Zero-1-to-3: Zero-shot One Image to 3D Object," in IEEE/CVF International Conference on Computer Vision (ICCV), Paris, France, pp. 9298-9309, 2023, DOI: 10.1109/ICCV51070.2023.00853.
  11. X. Long, Y.-C. Guo, C. Lin, Y. Liu, Z. Dou, L. Liu, Y. Ma, S.-H. Zhang, M. Habermann, C. Theobalt, and W. Wang, "Wonder3D: Single Image to 3D using Cross-Domain Diffusion," in IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Seattle, WA, USA, pp. 9970-9980, 2024, DOI: 10.1109/CVPR52733.2024.00951.
  12. Y. Hong, Y. Zhang, C. Pan, Z. Yang, and L. Liu, "LRM: Large Reconstruction Model for Single Image to 3D," in The Twelfth International Conference on Learning Representations (ICLR), Vienna, Austria, 2024, [Online], https://openreview.net/forum?id=sllU8vvsFF.
  13. B. Poole, A. Jain, J. T. Barron, and B. Mildenhall, "DreamFusion: Text-to-3D using 2D Diffusion," in The Eleventh International Conference on Learning Representations (ICLR), Kigali, Rwanda, 2023, [Online], https://openreview.net/forum?id=FjNys5c7VyY.
  14. E. J. Hu, Y. Shen, P. Wallis, Z. Allen-Zhu, Y. Li, S. Wang, L. Wang, and W. Chen, "LoRA: Low-Rank Adaptation of Large Language Models," in The Tenth International Conference on Learning Representations (ICLR), Virtual, 2022, [Online], https://openreview.net/forum?id=nZeVKeeFYf9.
  15. Black Forest Labs, Introducing FLUX.2, [Online], https://bfl.ai/blog,Accessed: Mar. 11, 2026.
  16. Black Forest Labs, FLUX.2-klein-9B, Hugging Face Model Card, [Online], https://huggingface.co/black-forest-labs/FLUX.2-klein9B, Accessed: Mar. 11, 2026.
  17. T. Brooks, A. Holynski, and A. A. Efros, "InstructPix2Pix: Learning To Follow Image Editing Instructions," in IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Vancouver, Canada, pp. 18392-18402, 2023, DOI: 10.1109/CVPR52729.2023.01764.
  18. Z. Ge, S. Liu, F. Wang, Z. Li, and J. Sun, "YOLOX: Exceeding YOLO Series in 2021," arXiv:2107.08430, 2021, DOI: 10.48550/arXiv.2107.08430.
  19. P. Lindenberger, P.-E. Sarlin, and M. Pollefeys, "LightGlue: Local Feature Matching at Light Speed," in IEEE/CVF International Conference on Computer Vision (ICCV), Paris, France, pp. 17627-17638, 2023, DOI: 10.1109/ICCV51070.2023.01616.
  20. P. C. Lusk, K. Fathian, and J. P. How, "CLIPPER: A GraphTheoretic Framework for Robust Data Association," in 2021 IEEE International Conference on Robotics and Automation (ICRA), Xi'an, China, pp. 13828-13834, 2021, DOI: 10.1109/ICRA48506.2021.9561069.