The KITTI Vision Benchmark Suite

In this paper, we propose a novel three-branch
backbone comprising color-guided, semantic-guided,
and depth-guided branches. Specifically, the
color-guided branch takes a sparse depth map and
RGB image as an input and generates color depth
which includes color cues (e.g., object
boundaries) of the scene. The predicted dense
depth map of color-guided branch along-with
semantic image and sparse depth map is passed as
input to semantic-guided branch for estimating
semantic depth. The depth-guided branch takes
sparse, color, and semantic depths to generate the
dense depth map. The color depth, semantic depth,
and guided depth are adaptively fused to produce
the output of our proposed three-branch backbone.
In addition, we also propose to apply semantic-
aware multi-modal attention-based fusion block
(SAMMAFB) to fuse features between all three
branches. We further use CSPN++ with Atrous
convolutions to refine the dense depth map
produced by our three-branch backbone.

lr=0.0012, bs = 8

@ARTICLE{9918022, author={Nazir, Danish and Pagani,
Alain and Liwicki, Marcus and Stricker, Didier and
Afzal, Muhammad Zeshan}, journal={IEEE Access},
title={SemAttNet: Towards Attention-based Semantic
Aware Guided Depth Completion}, year={2022},
volume={}, number={}, pages={1-1}, doi=
{10.1109/ACCESS.2022.3214316}}