installing and testing pyslam under Windows

[Luigi Freda]

Many people asked me how to install and test pyslam under Windows.
I would suggest going with Windows WSL2 and install ubuntu under it.
Didn't test it yet but I found this cool install procedure that should provide a fast ubuntu GUI under Windows
https://www.youtube.com/watch?v=IL7Jd9rjgrM

WIP ...

[Xiao Hu]

Got this repo to work flawlessly with WSL2 under Windows 10 and Mac OS Big Sur!! Thank you for the great documentation!

[Luigi Freda]

Awesome! Thanks for sharing your experience!

[Xiao Hu]

No problem!! Just so you know, there was one hiccup that I encountered during the building of g2opy regarding the Eigen quaternion member access. I followed this pull request of g2opy official repo and got it to work. Maybe you can include this in your README in case anyone else has similar problems. 😄

[Luigi Freda]

Thanks for the feedback!
That hiccup is really strange! I've managed that problem here:
https://github.com/luigifreda/pyslam/blob/ubuntu20/thirdparty/g2opy_changes/eigen_types.h#L185
Did you use the ubuntu20 branch? (This is the branch that must be used under Ubuntu 20 as explained in the README)

[Luigi Freda]

I've just checked again that, both with docker and a real Ubuntu 20 install, g2opy get compiled correctly under the ubuntu20 branch.

[Zhang Xianglong]

I run pySLAM under win10, and only run the visual odometer that is main_vo.py. The trajectory error on the Kitti data set is very small, as shown in the first figure, but the results are basically incorrect when running the TUM data set, as shown in the second as shown in the figure. I have changed the camera parameters in the program, why is this? I have tried several TUM datasets and they are all wrong~ I can't find where the problem is~

[Luigi Freda]

Hi,
you can't use main_vo.py on TUM datasets.
This is well explained in the README:

Indeed, with main_vo.py

you need to use a ground truth in order to recover a correct inter-frame scale s and estimate a valid trajectory by composing $C_k = C_{k-1} * [R_{k-1,k}, s t_{k-1,k}]$.
This script is a first start to understand the basics of inter-frame feature tracking and camera pose estimation.

You need the full main_slam.py in order to run other datasets.

You can find further details in this related thread #20.
As a side note, I think it's always great to take some time to read the main README file 🙂

[Zhang Xianglong]

Hi, you can't use main_vo.py on TUM datasets. This is well explained in the REAMDE

Indeed, with main_vo.py

you need to use a ground truth in order to recover a correct inter-frame scale s and estimate a valid trajectory by composing $C_k = C{k-1} * [R_{k-1,k}, s t_{k-1,k}]$._ This script is a first start to understand the basics of inter-frame feature tracking and camera pose estimation.

You need the full main_slam.py in order to run other datasets.

Thank you so much for your guidance, this pySLAM project of yours is really nice.

[Zhang Xianglong]

Hello, I used main_slam.py to run the TUM dataset, but I only want the function of visual odometry, no need to map, how should I change it in the pySLAM program? Or what features are turned off?

[Luigi Freda]

Hello @zxl152723,
Did you have time to read the main README file?
Did you study the code?
What's your experience with SLAM?