Cameras file, ORB -- Sift. Strange!

WebODM
#1

I’ve took a part of my dataset, a part where two flights overlaped, and run both Sift and ORB to look at how different the camera would be interpreted.

It was a bit shocking to see the “k2” value be so different between the two.

Is this normal?

Sift:
“dji fc6310r 5472 3648 fisheye 0.6666”: {
“projection_type”: “fisheye”,
“width”: 5472,
“height”: 3648,
“focal”: 0.6615437546312625,
“k1”: 0.04734803759865527,
“k2”: 3.906486909854561e-05
}

ORB:
“dji fc6310r 5472 3648 fisheye 0.6666”: {
“projection_type”: “fisheye”,
“width”: 5472,
“height”: 3648,
“focal”: 0.666329134203784,
“k1”: 0.042137408948134926,
“k2”: 0.0027238883685679004
}

#2

I’m not shocked at all :slightly_smiling_face:
Orb uses the central part and SIFT covers the whole field of view. You have fisheye projection selected, and the image scale on a fisheye lens varies hugely from centre to edge, distorting and compressing more and more towards the edges. SIFT has to correct for that outer edge distortion, ORB doesn’t to anywhere near as much.

1 Like
#3

But how does it affect reconstruction?

#4

I don’t know, can you post the orthos so we can see any differences?

#5

I’m more interested in point clouds so I don’t compute an ortophoto

#6

Pointcloud looks much better using the cameras file from a sift process in a orb process.

#7

This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.