The KITTI Vision Benchmark Suite

This paper proposed an efficient real-time
semantic segmentation model, called Short-term
Dense Bottleneck Network (SDBNet). It introduces
two different types of shoft-term dense bottleneck
blocks which provides diverse field-of-view for
better contextualization of varied geometrical
objects in a complex scene. This model is
evaluated by three publicly available benchmarks
and produced a competitive performance while
having only 1.45 million parameters and 9.9 GFLOPs
at KITTI image resolution.

SDBNet has only 1.45 million parameters.

@INPROCEEDINGS{singha2022sdbnet,
author={Singha, Tanmay and Pham, Duc-Son and
Krishna, Aneesh},
booktitle={Proc. DICTA},
title={{SDBNet: Lightweight Real-Time Semantic
Segmentation Using Short-Term Dense Bottleneck}},
year={2022},
volume={},
number={},
pages={1-8},
doi={10.1109/DICTA56598.2022.10034634}}