ViT在工业密集预测中数据效率高，源于预训练一致性而非注意力机制优势。
Rethinking Data Efficiency in Industrial Dense Prediction: Pretraining Coherence, Not Inductive Bias, Determines ViTs Low-Data Advantage
- 通过特征对齐块修复ViT与CNN预训练不一致问题。
- 200样本以上时Swin-Graft mAP达0.973，优于YOLOv11x。
- 适合解决工业场景小样本密集预测任务的研究者参考。
视觉变换器（ViTs）常被认为在工业密集预测任务中需要比卷积神经网络（CNNs）更多的标注数据。通过对四个工业数据集的受控实验，我们发现数据效率差距的根源在于预训练不一致：ImageNet预训练的ViT主干与COCO预训练的CNN颈部之间存在统计失配，而非自注意力机制本身缺陷。我们量化了跨架构特征差异，并提出轻量级AlignBlock家族实现金字塔级特征重校准。核心发现揭示了数据效率前沿：在领域相近场景中，样本数≥200时，Swin-Graft在703张图像终端测试下达到0.973 mAP@50，超过YOLOv11x的0.956；而在领域相距较远场景中，CNN仍具优势（141张图像时为0.900 vs 0.600 mAP@50）。嫁接颈部权重可使性能提升至随机初始化颈部的2.5倍。
原文摘要 · Abstract (English)
Vision Transformers (ViTs) are widely believed to require more labeled data than CNNs for industrial dense prediction. Through controlled experiments on four industrial datasets, we show that the data-efficiency gap stems from pretraining incoherence, which refers to the statistical mismatch between ImageNet-pretrained ViT backbones and COCO-pretrained CNN necks, rather than from inherent self-attention deficits. We characterize the cross-architecture feature gap and propose a lightweight AlignBlock family for pyramid-level feature recalibration. Our core finding empirically identifies a data-efficiency frontier: for domain-proximal scenes with >= 200 samples, Swin-Graft surpasses YOLOv11x (terminal 703-shot: 0.973 vs 0.956 mAP@50); for domain-distant scenes, CNNs retain advantage (hook 141-shot: 0.900 vs 0.600 mAP@50). Grafted neck weights yield up to 2.5x the mAP of a randomly initialized neck.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。