用高质量合成数据和分层去噪，让检测模型在少量训练后就能准确识别真实图像中的物体。
High-Quality Proposal Encoding and Cascade Denoising for Imaginary Supervised Object Detection
- 用LLaMA-3、Flux和Grounding DINO构建高质合成数据集，实现从弱监督到全监督的提升。
- 通过图像特定先验初始化查询，加速收敛并避免对合成数据过拟合。
- 分层动态去噪机制按层级提高匹配阈值，引导模型学习鲁棒边界而非噪声标签。
目标检测模型依赖大规模标注数据，但人工标注成本高。这催生了想象式监督检测（ISOD），即模型在合成图像上训练，在真实图像上测试。然而现有方法存在三方面局限：（1）合成数据提示简单、图像质量差、监督弱；（2）基于DETR的检测器因随机查询初始化导致收敛慢，易过拟合合成模式，影响真实泛化能力；（3）统一去噪压力促使模型过度适应伪标签噪声。本文提出级联高质量提议-去噪DETR（Cascade HQP-DETR）来解决上述问题。首先，利用LLaMA-3、Flux与Grounding DINO构建高质数据流水线，生成FluxVOC与FluxCOCO数据集，推动ISOD从弱监督迈向全监督。其次，引入高质量提议引导的查询编码机制，以SAM生成的提议和RoI池化特征作为图像特定先验初始化对象查询，加速收敛并引导模型学习可迁移特征，避免过拟合合成模式。第三，提出级联去噪算法，通过逐层递增的IoU阈值动态调整训练权重，引导模型从可靠视觉线索中学习鲁棒边界，而非盲目拟合噪声标签。仅在FluxVOC上训练12个周期，该模型在PASCAL VOC 2007上达到61.04% [email protected]，超越多个强基线，其在真实数据上的优异表现验证了架构的通用性。
原文摘要 · Abstract (English)
Object detection models demand large-scale annotated datasets, which are costly and labor-intensive to create. This motivated Imaginary Supervised Object Detection (ISOD), where models train on synthetic images and test on real images. However, existing methods face three limitations: (1) synthetic datasets suffer from simplistic prompts, poor image quality, and weak supervision; (2) DETR-based detectors, due to their random query initialization, struggle with slow convergence and overfitting to synthetic patterns, hindering real-world generalization; (3) uniform denoising pressure promotes model overfitting to pseudo-label noise. We propose Cascade HQP-DETR to address these limitations. First, we introduce a high-quality data pipeline using LLaMA-3, Flux, and Grounding DINO to generate the FluxVOC and FluxCOCO datasets, advancing ISOD from weak to full supervision. Second, our High-Quality Proposal guided query encoding initializes object queries with image-specific priors from SAM-generated proposals and RoI-pooled features, accelerating convergence while steering the model to learn transferable features instead of overfitting to synthetic patterns. Third, our cascade denoising algorithm dynamically adjusts training weights through progressively increasing IoU thresholds across decoder layers, guiding the model to learn robust boundaries from reliable visual cues rather than overfitting to noisy labels. Trained for just 12 epochs solely on FluxVOC, Cascade HQP-DETR achieves a SOTA 61.04\% [email protected] on PASCAL VOC 2007, outperforming strong baselines, with its competitive real-data performance confirming the architecture's universal applicability.
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。