sxyu/rgbdrec

Depth camera pose estimation utility, with common abstraction on top of COLMAP, ORB_SLAM2

Website: https://alexyu.net

Discord: @doriath0

rgbdrec's Issues

The transform from x -y -z to x y z

I cannot get the idea of these code. Could you please explain in detail:
// x -y -z -> x y z
trans.bottomRightCorner<2, 1>() *= -1;
trans.block<1, 2>(0, 1) *= -1;
trans.bottomLeftCorner<2, 1>() *= -1;

In general, the reflection matrix or rotation 180 degree along x is :
1 0 0
0 -1 0
0 0 -1

Increase the accuracy of the point cloud(intrinsic and extrinsic parameters)

Thank you very much for sharing this.

i use the rgbdreg-colmap. But the result i obtained for my data set is not that good.
I got this as well :WARNING: scale error is high, poses may be too inaccurate.
Is there a way that i can increase the accuracy of the point cloud (intrinsic and extrinsic parameters) in this implementation ?

I need to register the point cloud from rbgd images.Currenly i am doing it with orb slam2. It has a considerable accuracy. But now i need to increase the accuracy of the registered point cloud (increase the accuracy of the intrinsic and extrinsic parameters).That is why i tired your method.

Can u suggest any method which can increase the accuracy of the point cloud which obtained from the orb slam2 or any new method that i can use to achieve it ?

imagesFs.read(reinterpret_cast<char*>(q.vec().data()), sizeof(double) * 3);

q has 4 members, why read 3 doubles?
In rgbdreg-colmap.cpp lines 162-163:
imagesFs.read(reinterpret_cast<char*>(q.vec().data()),
sizeof(double) * 3);