The KITTI Vision Benchmark Suite

Monocular 3D object detection has garnered
widespread attention in the field of autonomous
driving due to its advantages of low cost and low
time consumption. Current research is primarily
focused on the independent detection of targets,
with the accuracy significantly compromised in the
presence of some occluded targets in the scene.
Improving the accuracy of detecting occluded
targets is a critical and urgent challenge. We
argue that the use of local visual or depth
features alone does not fully characterize the
scene-level 3D spatial structure because it
ignores the implicit positional relationships
between instances in the scene. In this paper, we
propose to use high-precision 2D center
information to guide depth information in order to
obtain the implicit spatial constraints between
occluded and unoccluded targets, called Occlude3D.
Specifically, we design the Center-Feature Fusion
Module (CFM) and Decoupled Depth-Feature Module
(DDFM) to process the input visual feature and
depth

\

\