This paper introduces a novel approach for learning depth and surface normal from monocular images by incorporating geometric context. The key idea is to adaptively determine the faithful local geometry indicated by the learned geometric context to correlate depth and surface normal. The method first produces a geometric context map that encodes 3D geometric variances, then jointly predicts depth and surface normal in a geometry-aware manner. The geometric context is used to enforce an adaptive surface normal constraint on the predicted depth, enabling the depth to faithfully preserve 3D geometry. The surface normal is estimated guided by the geometric context, capturing rich geometric details. The method unifies depth and surface normal estimation within a cohesive framework, enabling high-quality 3D geometry generation from images. The approach is validated on diverse indoor and outdoor datasets, showing superior performance compared to state-of-the-art methods. The method is efficient and robust, with code and data available at <https://www.xxlong.site/ASNDepth/>. The paper also discusses related work, including monocular depth estimation, joint depth and normal estimation, and edge preserving methods. The method is evaluated on NYUD-V2, ScanNet, MVS-SYNTH, and SVERS datasets, showing significant improvements in depth estimation, surface normal estimation, and point cloud quality. The method is also analyzed through ablation studies and visualizations, demonstrating its effectiveness in capturing geometric details and robustness to noise and variations.This paper introduces a novel approach for learning depth and surface normal from monocular images by incorporating geometric context. The key idea is to adaptively determine the faithful local geometry indicated by the learned geometric context to correlate depth and surface normal. The method first produces a geometric context map that encodes 3D geometric variances, then jointly predicts depth and surface normal in a geometry-aware manner. The geometric context is used to enforce an adaptive surface normal constraint on the predicted depth, enabling the depth to faithfully preserve 3D geometry. The surface normal is estimated guided by the geometric context, capturing rich geometric details. The method unifies depth and surface normal estimation within a cohesive framework, enabling high-quality 3D geometry generation from images. The approach is validated on diverse indoor and outdoor datasets, showing superior performance compared to state-of-the-art methods. The method is efficient and robust, with code and data available at <https://www.xxlong.site/ASNDepth/>. The paper also discusses related work, including monocular depth estimation, joint depth and normal estimation, and edge preserving methods. The method is evaluated on NYUD-V2, ScanNet, MVS-SYNTH, and SVERS datasets, showing significant improvements in depth estimation, surface normal estimation, and point cloud quality. The method is also analyzed through ablation studies and visualizations, demonstrating its effectiveness in capturing geometric details and robustness to noise and variations.