Research map / Detection and segmentation / 3D detection and driving

Rethinking Geometric Depth in Monocular 3D Object Detection: A Projection-Consistent Reformulation

NeurIPS 2026 · Detection and segmentation › 3D detection and driving · approach: Monocular 3D object detection

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Shows that the geometric depth formula z ≈ f·H/h_2D carries a projection-induced bias, recovering the visible front surface rather than the object center, and proposes a projection-consistent reformulation for monocular 3D detection.

Problem

Monocular 3D object detection relies on recovering depth from a single image; the widely used geometric depth formula z ≈ f·H/h_2D is simple but may be systematically biased.

Approach

Identifies that substituting the 2D box height into the geometric formula yields the depth of the object's visible front surface rather than its center, and reformulates depth estimation to be consistent with the projection model.

Surprising findings

• The standard geometric depth formula, using the 2D box height, systematically recovers the depth of the visible front surface rather than the object center.

Closest papers by meaning

• MonoDGP: Monocular 3D Object Detection with Decoupled-Query and Geometry-Error Priors (CVPR 2025 · similarity 0.94)
• RefineAny3D: Depth Refinement as Semantic Alignment for Monocular 3D Detection (NeurIPS 2026 · similarity 0.91)
• MonoCD: Monocular 3D Object Detection with Complementary Depths (CVPR 2024 · similarity 0.90)
• MonoDiff: Monocular 3D Object Detection and Pose Estimation with Diffusion Models (CVPR 2024 · similarity 0.90)
• MonoPlace3D: Learning 3D-Aware Object Placement for 3D Monocular Detection (CVPR 2025 · similarity 0.90)
• MonoMAE: Enhancing Monocular 3D Detection through Depth-Aware Masked Autoencoders (NeurIPS 2024 · similarity 0.89)