Monocular 3D object detection has gained significant attention due to its potential to accurately localize objects in 3D from a single image at a low cost. Depth estimation is a crucial but challenging subtask in this field, often leading to ill-posed 2D to 3D mapping. Existing methods typically ensemble multiple local depth clues to improve accuracy, but these often have similar error signs, hindering the overall performance. To address this, the authors propose two novel designs to enhance the complementarity of depths: adding a new depth prediction branch that utilizes global and efficient depth clues from the entire image, and exploiting geometric relations between multiple depth clues to achieve complementary form. Their method, named MonoCD, is evaluated on the KITTI benchmark and demonstrates state-of-the-art performance without introducing extra data. Additionally, the complementary depth can be used as a lightweight module to boost existing monocular 3D object detectors. The paper also includes a detailed analysis of the effectiveness of complementary depths and their impact on detection accuracy, as well as qualitative and quantitative results to support the claims.Monocular 3D object detection has gained significant attention due to its potential to accurately localize objects in 3D from a single image at a low cost. Depth estimation is a crucial but challenging subtask in this field, often leading to ill-posed 2D to 3D mapping. Existing methods typically ensemble multiple local depth clues to improve accuracy, but these often have similar error signs, hindering the overall performance. To address this, the authors propose two novel designs to enhance the complementarity of depths: adding a new depth prediction branch that utilizes global and efficient depth clues from the entire image, and exploiting geometric relations between multiple depth clues to achieve complementary form. Their method, named MonoCD, is evaluated on the KITTI benchmark and demonstrates state-of-the-art performance without introducing extra data. Additionally, the complementary depth can be used as a lightweight module to boost existing monocular 3D object detectors. The paper also includes a detailed analysis of the effectiveness of complementary depths and their impact on detection accuracy, as well as qualitative and quantitative results to support the claims.