Based on the LSS (Lift, Splat, Shoot) algorithm, an extension was made to simultaneously detect vehicles, vehicle-road interactions, lane lines, and segmentation parts of the road, resulting in slightly improved performance compared to what was reported in the paper.
基于LSS (Lift, Splat, Shoot)算法,增加了同时检测车辆、道路与车辆、车道线、道路的分割部分。并获得了比论文中提到的稍好的效果。为个人本科毕业设计。
git clone https://github.com/byx020119/LSS_reproduction.git
cd LSS_reproduction
python main.py viz_model_preds mini --modelf=MODEL_ROOT --dataroot=NUSCENES_ROOT --map_folder=MAP_FOLDER_ROOT --outC=CLASS_NUMBER --gpuid=GPU_ID
annotations: outC=class number
default:
- outC=1:vehicle
- outC=2:vehicle and road
- outC=3:vehicle, divider and road
@inproceedings{philion2020lift,
title={Lift, Splat, Shoot: Encoding Images From Arbitrary Camera Rigs by Implicitly Unprojecting to 3D},
author={Jonah Philion and Sanja Fidler},
booktitle={Proceedings of the European Conference on Computer Vision},
year={2020},
}