using stereo fisheye based on this project

[ns15417]

Hi, thanks for your project and i already got it running on my computer very well. and I add some code based on the idea of multicol fisheye slam and now running on stereo state to fix the scale problem of mono. As i tried, the stereo works well in the room but failed if i go to another room and back again. By "fail", i mean the go and back path can not overlapped together even after global BA. and also the global BA is quiet slow ,sometimes it causes 10 mins.

Have you tried about the stereo one and do you have any idea or suggestion about my problem?
I attached my running result here to help the understanding of my problem.
Indoor result（one room）:

two room：

[lsyads]

Hi. When run in a monocular camera, the global BA is also slow in the large environment. I guess the fisheye camera make too many active points, so the computation cost is very big. The active points maybe need to be reduced. I'm sorry that I didn't try about the stereo one.

[engcang]

@ns15417 Do you have source code of your stereo project? It sounds really interesting.

[ns15417]

@ns15417 Do you have source code of your stereo project? It sounds really interesting.

Hi, i do have the code but it is still on stage and not very stable. So i am afraid cannot share it now.
But actually what i did is only on the g2o optimization part and the Check3D() function in Frame, change the code by those functions and it works in small rooms.