文章目录
提示:整理了感知算法中性能优异的模型
1. 激光雷达目标检测(WXF)
-
SFD
代码:https://github.com/LittlePey/SFD
论文:https://xueshu.baidu.com/usercenter/paper/show?paperid=123q0xs0a3010an0qh1w0mn0xj564930 -
BtcDet
代码:https://github.com/Xharlie/BtcDet
论文:https://arxiv.org/pdf/2112.02205.pdf -
SE-SSD
代码:https://github.com/Vegeta2020/SE-SSD
论文:https://arxiv.org/abs/2104.09804 -
Focals Conv
代码:https://github.com/dvlab-research/FocalsConv
论文:https://arxiv.org/abs/2204.12463
2.激光雷达语义分割(GNN)
- SPVCNN++ ,SPVNAS
代码:https://github.com/mit-han-lab/spvnas
论文:https://arxiv.org/pdf/2007.16100.pdf
- Cylinder3D++, Cylinder3D
代码:https://github.com/xinge008/Cylinder3D
论文:https://arxiv.org/pdf/2011.10033.pdf
-
JS3C-Net
代码:https://github.com/yanx27/JS3C-Net
论文:https://arxiv.org/abs/2012.03762 -
Vis-PolarNet、PolarNet
代码:https://github.com/AbangLZU/PolarSeg.git
https://github.com/edwardzhou130/PolarSeg
论文:https://arxiv.org/abs/2003.14032
训练记录博客。
3. 视觉目标检测(PYL)
-
swin transformer v2
coco数据集上的box-map(目标检测方面)是63.1。
代码:https://github.com/microsoft/Swin-Transformer
论文:https://arxiv.org/pdf/2111.09883v1.pdf -
yolov5x6
在coco数据集上的map是55.8。
代码:https://github.com/ultralytics/yolov5/releases
论文:暂无。
4. 姿态识别(ZCX)
-
SOTA:Soft-gated Skip Connections
paper: https://arxiv.org/pdf/2002.11098v1.pdf
code: https://github.com/salinasJJ/BBpose -
SOTA: ViTPose(ViTAE-G, ensemble):
paper: https://arxiv.org/pdf/2D2D0K4.12D484v2D.pdf
code: https://github.com/vitae-transformer/vitpose -
SOTA: ViTAE:
paper: https://arxiv.org/pdf/2204.12484v2.pdf
code: https://github.com/vitae-transformer/vitpose
姿态识别数据集、模型介绍详细见文中第5节内容:6. 详细解释–Pose Estimate 榜单
5. 车道线识别(YGD)
车道线检测数据集、模型介绍详细见文中第6节内容:7. 详细解释–车道线检测SOTA模型
4、5节的详细解释部分:
6. 详细解释第4节–Pose Estimate 榜单
7. 详细解释第5节–车道线检测SOTA模型