The KITTI Vision Benchmark Suite

In this work, we present DUO-Net, a unified multi-
task learning framework for joint 2D object
detection and depth estimation. The architecture
employs a shared ResNet-based backbone with
attention modules and task-specific heads to
simultaneously perform bounding box localization
and dense depth prediction. We adopt a two-stage
training strategy involving task-specific
pretraining followed by joint optimization to
enhance representation learning and stability.

alpha=0.0

@inproceedings{ghaffar2025duo,
title={DUO-Net: Joint End-to-End 2D Object
Detection and Depth Estimation via Uncertainty-
Aware Multitask Learning},
author={Ghaffar, Fazal and Khan, Burhan and
Jalali, Seyed Mohammad J and Lim, Chee Peng},
booktitle={2025 IEEE International Conference on
Systems, Man, and Cybernetics (SMC)},
pages={4377--4384},
year={2025},
organization={IEEE}
}