search

sair-lab / iSLAM

iSLAM: Imperative SLAM (RA-L 2024) is a novel Visual-Inertial SLAM using Self-supervised Learning
https://sairlab.org/iSLAM/
Other
113 stars 9 forks source link

How do you define the *low* vs *high* level? #2

Closed nnop closed 2 months ago

nnop commented 9 months ago

Great idea for self-supervised SLAM! I have a question about how do you define the low vs high for level. Is it because BA optimize against the raw image observation, while deep networks learns the high level feature?

But in another perspective, BA seems more high level for it's closer to the final purpose, while feature matching is more low level for it's doing raw image feature matching.

Hope to know about your definitions for this. @wang-chen

wang-chen commented 9 months ago

Hi, a high-level answer is: an upper-level optimization targets tasks that are suitable for learning-based methods, while low-level optimization targets tasks that are suitable for geometry-based methods.

FuTaimeng commented 2 months ago

Thanks for your interest in our work!

Since there haven't been any new comments for a while, I'll close this issue.