LAF-YOLOv10：无人机航拍图像中的小目标检测
Small Object Detection in Drone Aerial Imagery with LAF-YOLOv10
浏览论文内容
中文总结 AI 辅助
针对无人机小目标检测，LAF-YOLOv10 集成四种技术于 YOLOv10n，实验发现组合效果并非各技术单独效果的简单叠加，P2/P5 头与骨干交互导致性能下降，强调可组合性需直接验证。
中文摘要 AI 辅助
通用目标检测器在无人机镜头上的精度会下降，因为目标仅覆盖少量像素，且机载计算能力有限。先前的工作将独立验证的架构技术组合成一个检测器，假设单独报告的性能提升在组合后仍然有效。我们直接检验了这一假设。LAF-YOLOv10 将四种技术集成到 YOLOv10n 中：部分卷积 C2f（PC-C2f）骨干模块、注意力引导的特征金字塔网络（AG-FPN）、替换大目标 P5 头的 P2 检测头，以及 Wise-IoU v3 回归损失，探究它们的组合效果是否与各自单独贡献相匹配。我们在 VisDrone-DET2019 上使用三个随机种子（42、123、256）训练 LAF-YOLOv10 三次，与未修改的 YOLOv10n 进行基准比较，并使用 TIDE 错误分解、逐类别分析、逐组件消融、注意力/损失比较、对 UAVDT 的零样本迁移以及留出集/测试开发集评估，来定位组合成功或失败的具体环节。可组合性在此并不成立。LAF-YOLOv10 在 2.14M 参数下达到 24.0±0.4% 的 mAP@0.5，比 YOLOv10n（31.8%）低 7.8 个百分点，这一差距也迁移到了 UAVDT（-10.0 个百分点），并得到留出集和测试开发集评估（23.5%、22.5%）的确认。背景误检、定位误差和重复检测的变化方向与 AG-FPN 和 Wise-IoU 的设计初衷一致。消融实验将性能下降追溯到具体来源：P2/-P5 头替换独立造成 2.5 个百分点的损失，加上与 PC-C2f 削弱的骨干叠加时的 2.5 个百分点交互惩罚，而 PC-C2f 自身的 2.0 个百分点损失与部分预训练权重移植（73/150 个骨干张量迁移）一致。失败归因于特定交互，而非四个组件本身。可组合性必须直接验证，而不能假设。代码/检查点：此 https URL。
英文摘要
General-purpose object detectors lose accuracy on UAV footage, where targets span only a handful of pixels and onboard compute is limited. Prior work composes independently-validated architectural techniques into one detector, assuming gains reported in isolation transfer once combined. We stress-test that assumption directly. LAF-YOLOv10 integrates four techniques into YOLOv10n: a Partial Convolution C2f (PC-C2f) backbone block, an Attention-Guided Feature Pyramid Network (AG-FPN), a P2 detection head replacing the large-object P5 head, and Wise-IoU v3 regression, asking whether their combined effect matches what each contributes alone. We train LAF-YOLOv10 three times (seeds 42, 123, 256) on VisDrone-DET2019, benchmark against unmodified YOLOv10n, and use TIDE error decomposition, per-category breakdown, per-component ablation, attention/loss comparisons, zero-shot transfer to UAVDT, and held-out/test-dev evaluation to localize where the combination succeeds or fails. Composability does not hold here. LAF-YOLOv10 reaches 24.0+/-0.4% mAP@0.5 at 2.14M parameters, 7.8 points below YOLOv10n (31.8%), a deficit that transfers to UAVDT (-10.0 points) and is confirmed by held-out and test-dev evaluation (23.5%, 22.5%). Background false positives, localization error, and duplicate detections move in the direction AG-FPN and Wise-IoU were designed to push. Ablation traces the deficit to a specific source: the P2/-P5 head swap costs 2.5 points independently plus a 2.5-point interaction penalty when layered onto a backbone weakened by PC-C2f, whose own 2.0-point loss is consistent with a partial pretrained-weight transplant (73/150 backbone tensors transfer). The failure is attributable to a specific interaction, not the four components individually. Composability must be verified directly, not assumed. Code/checkpoints: https://github.com/Mudassiruddin7/Small-Object-Detection-in-UAV-Imagery.
