Multiple errors with colmap cuda extraction and others

[Jaykumaran]

Hello,

1. I'm using colmap feature handler and i'm facing multiple errors w.r.t rotation avg, track establishment, and others
   For eg,

i)
File "/media/instantsfm/instantsfm/processors/track_establishment.py", line 32, in BlindConcatenation
point_global_id2 = (int(pair.image_id2) << 32) | point2
OverflowError: Python integer 4294967296 out of bounds for uint32

ii)
return forward_call(*args, **kwargs)
TypeError: RobustModel.forward() missing 1 required positional argument: 'target'

2. Also im trying superpoint, disk+lightglue etc and the ins-feat for this is killed when it goes out of ram (i..e 64 + 128 swap) which doesnt make it practical for large scenes. Can someone guide on these deep learning feature extractor to get accurate poses and Sfm for large scenes

[cre185]

Hi,

1. From the error it seems an enormous number of matching points are calculated? I wonder if you could provide some extra information, i.e. the processing output or numbers of matching points in your database.db?
2. Feature handlers except for colmap indeed have their problems and are still not officially supported. We're planning to incorporate more possible features into our pipeline, but currently only colmap is available due to its stability and robustness.

[Jaykumaran]

Hi,

1. I'm trying a custom drone dataset which moves in linear fashion. Yes the resolution is high and i'm processing nearly 1900 images
2. Thanks for the clarification

[cre185]

Then I think resolution can indeed be a problem. You can try to downsample your images before feeding them into colmap's feature extractor.