prbonn / overlap_localization Goto Github PK

Hi, thanks for your great work!
I just look at the localization result gif you provide, the location error drops dramatically from more than 4 to a fairly low number at the first few frames, while the heading error stays at a normal level, what happened during this period?
Thank you.

Hi @Chen-Xieyuanli Thanks a lot for your open-sourced code!

I found some minor difference between the .bin scan files from your compressed zip file and the original KITTI odometry dataset scans. I'm opening this issue for just letting you know and check.

Happy New Year!

Best,
XIn

Hi @Chen-Xieyuanli . Sorry to bother you. It's me again [: )]

I found the provided model in the zip file. It seems the model is trained on KITTI seq 07 with a train/val split. I know it's a toy example for users to understand how it works. I have two questions:

1. What should I do if I wish to evaluate overlap model's performance on KITTI fairly? I think one workable way is to train the model on other sequences and test it on seq 07.
2. How does the overlap model trained on single scans perform on a scan to local map test? I guess there exists some gap between "scan to scan" and "scan to local map".

I would be very appreciate if you can give me some advice. Thanks a lot!