这里重点是摄像头运动估计和定位，不是地图或者路标（landmark）。以前大家都知道SLAM结合深度学习最多的是语义SLAM，比如语义分割和语义目标识别。而这里强调的是里程计和定位。另外，忽略一些特征提取和匹配的方法。
1. DEMON
https://github.com/lmb-freiburg/demongithub.com
2. SfM Learner
https://github.com/tinghuiz/SfMLearnergithub.com
3. LEGO
zhenheny/LEGOgithub.com
4. Vid2Depth
https://sites.google.com/view/vid2depthsites.google.com
5. DeepMatchVO
https://github.com/hlzz/DeepMatchVOgithub.com
6. DeepVO
ChiWeiHsiao/DeepVO-pytorchgithub.com
krrish94/DeepVOgithub.com
7. DDVO
https://github.com/MightyChaos/LKVOLearnergithub.com
8. MonoDepth2
https://github.com/nianticlabs/monodepth2github.com
9. Depth VO Feat
Huangying-Zhan/Depth-VO-Featgithub.com
10. SC SfM Learner
https://github.com/JiawangBian/SC-SfMLearner-Releasegithub.com
11. GeoNet
https://github.com/yzcjtr/GeoNetgithub.com
12. Nvidia CC
anuragranj/ccgithub.com
13. DOP Learning
guangmingw/DOPlearninggithub.com
14. EPC
chenxuluo/EPCgithub.com
15. DF-VO
Huangying-Zhan/DF-VOgithub.com
16. struct2depth
https://sites.google.com/view/struct2depthsites.google.com
17. DF-Net
DF-Netyuliang.vision
18. Samsung Odometry
saic-vul/odometrygithub.com
19. SfM-Net
augustelalande/sfmgithub.com
20. CNN-SVO
https://github.com/yan99033/CNN-SVOgithub.com
21. DeepTAM
https://github.com/lmb-freiburg/deeptamgithub.com
22. Active Neural SLAM
https://github.com/devendrachaplot/Neural-SLAMgithub.com
误差比较表格
轨迹比较图
来源：计算机视觉深度学习和自动驾驶
作者：黄浴
