Method

Face-Aware Depth Correction for Monocular 3D Detection [FA-MonoCD]


Submitted on 11 Mar. 2026 08:03 by
Taewook Eum (SKKU)

Running time:0.15 s
Environment:GPU @ 2.5 Ghz (Python)

Method Description:
A post-processing method for monocular 3D object
detection
that corrects depth estimates using LiDAR point
clouds and
instance segmentation masks. A Dynamic Utility
Gate selects
which detections to correct based on distance,
LiDAR point
quality, and face visibility. Corrected depths are
computed
via Weighted Anchor using visible 3D box faces.
Base detector: MonoCD. Segmentation: YOLOv8.
Parameters:
tau=0.45
Latex Bibtex:

Detailed Results

Object detection and orientation estimation results. Results for object detection are given in terms of average precision (AP) and results for joint object detection and orientation estimation are provided in terms of average orientation similarity (AOS).


Benchmark Easy Moderate Hard
Car (Detection) 84.18 % 61.65 % 49.38 %
Car (Orientation) 84.07 % 61.51 % 49.25 %
Car (3D Detection) 14.65 % 10.08 % 7.90 %
Car (Bird's Eye View) 25.90 % 18.41 % 14.36 %
This table as LaTeX

2D object detection results.
This figure as: png eps txt gnuplot



Orientation estimation results.
This figure as: png eps txt gnuplot



3D object detection results.
This figure as: png eps txt gnuplot



Bird's eye view results.
This figure as: png eps txt gnuplot




eXTReMe Tracker