用几何监督提升CAD与图像对齐精度，实现秒级零样本定位。
SUFLECA: Scaling Up Feature Learning for CAD-to-image Alignment
- 基于归一化物体坐标监督，从多源数据学习几何感知特征。
- 在ScanNet25k上达32.8%类别/42.6%实例准确率，超越最先进零样本方法。
- 无需迭代优化，适合机器人与增强现实中的实时应用。
CAD-to-image对齐旨在从单张RGB图像中估计物体的9维位姿（旋转、平移与各向异性尺度），应用于机器人和增强现实。现有零样本方法依赖视觉基础模型匹配图像区域与CAD模型，但对应关系常受外观影响，或在遮挡及合成到真实域偏移下不可靠。为此，我们提出SUFLECA（Scaling Up Feature Learning for CAD-to-image Alignment），一种弱监督的零样本CAD对齐框架，包含两大贡献：第一，通过在覆盖12个真实与合成数据集的图像上使用归一化物体坐标（NOCs）监督，扩展预训练视觉表示的几何引导特征学习，获得跨域泛化的紧凑几何感知特征；第二，提出几何一致匹配算法，建立可靠CAD-to-image对应关系。二者结合可实现每实例亚秒级精准对齐，无需迭代优化。在ScanNet25k上，SUFLECA取得32.8%/42.6%类别/实例准确率，优于最强零样本基线9.7/12.5个百分点，且计算开销更小，并首次在该基准上超越已有位姿监督方法。代码已开源。
原文摘要 · Abstract (English)
CAD-to-image alignment aims to estimate an object's 9D pose (rotation, translation, and anisotropic scale) from a single RGB image, with applications in robotics and augmented reality. Recent zero-shot methods use vision foundation models to match image regions to CAD models; yet their correspondences are typically appearance-driven or unreliable under occlusion or synthetic-to-real domain shift. To address these limitations, we introduce SUFLECA (Scaling Up Feature LEarning for CAD-to-image Alignment), a weakly supervised framework for zero-shot CAD alignment with two key contributions. First, SUFLECA scales up geometry-grounded feature learning from pretrained visual representations through Normalized Object Coordinates (NOCs) supervision on images spanning up to 12 real and synthetic datasets, learning compact geometry-aware features that generalize across domains. Second, we propose a geometrically consistent matching algorithm that establishes reliable CAD-to-image correspondences. Together, these contributions enable accurate, sub-second alignment per object instance without iterative pose refinement. On ScanNet25k, SUFLECA achieves 32.8%/42.6% category/instance accuracy, outperforming the strongest zero-shot baseline by 9.7/12.5 percentage points with a smaller computational footprint, and for the first time on this benchmark, even surpassing existing pose-supervised methods. Code is available at: https://github.com/snt-arg/SUFLECA
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。