Research map / Detection and segmentation / Open-set and few-shot detection
Diagnosing Aerial-View Object Detectors with Foundational Image Generative Models
Stanislav Panev, Minhyek Jeon, Vaishnavi Khindkar, Ahish Deshpande, Celso de Melo, Shuowen Hu, Shayok Chakraborty, Fernando de la Torre
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| supplement training with small real urban/industrial(/desert) sets chosen by synthetic diagnosis | ViTDet-B trained on UGRC · Synthetic diagnostic set | AP50: 81.6% -> 95.0% (+13.4%) | Improved |
| targeted supplementation | Faster R-CNN R50 trained on UGRC · Synthetic diagnostic set | AP50: 80.9% -> 93.5% | Improved |
| targeted supplementation | Faster R-CNN R50 trained on DOTAv2 · Synthetic diagnostic set | AP50: 93.8% -> 94.7% | No clear effect |
| targeted supplementation | YOLOv8-M trained on UGRC · Synthetic diagnostic set | AP50: 94.2% -> 95.9% | Improved |
| use random (LINZ or COCO) data at same budget instead of targeted data (ablation) | Faster R-CNN and YOLOv8 trained on DOTAv2 · scene-wise AP gains | AP50 | Hurt |
| use targeted data instead of random data at same budget (ablation) | YOLOv8 trained on DOTAv2 · urban scenes | AP50 gain: up to +11.7% | Improved |
| random data for ViTDet (ablation) | ViTDet-B trained on DOTAv2 · scene-wise AP gains | AP50 | No clear effect |
| evaluate detectors on synthetic scenes by scene type | Faster R-CNN, YOLOv8, ViTDet on DOTAv2/LINZ/UGRC · Synthetic diagnostic set | scene-wise AP50 | Improved |
Closest papers by meaning
- SkyScenes: A Synthetic Dataset for Aerial Scene Understanding (ECCV 2024 · similarity 0.88)
- AerialMegaDepth: Learning Aerial-Ground Reconstruction and View Synthesis (CVPR 2025 · similarity 0.86)
- Compositional Scene Understanding through Inverse Generative Modeling (ICML 2025 · similarity 0.85)
- Zero-Shot Aerial Object Detection with Visual Description Regularization (AAAI 2024 · similarity 0.84)
- Increasing the Utility of Synthetic Images through Chamfer Guidance (NeurIPS 2025 · similarity 0.84)
- GenRec: Knowing Where to Reconstruct and Where to Generate (NeurIPS 2026 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
