Impact of EXIF data on Point cloud


#22

Here’s the proper patch: https://github.com/mapillary/OpenSfM/commit/b27d7e142ce6a76d74aa273c802bf1deab68b83e?diff=split

I’ll get this into master. Just need to test whether the rest of OpenSfM is ok to come along too… .

@garlac – are you running as docker or natively?


#23

I am running it natively.

just now checked new file by replacing the old match_features.py with the new one provided in latest commit from opensfm.

I again got two partial reconstructions.Will test again if i missed anything.

Thank you.


#24

I’m only seeing one on my end. Which files are you looking at?


#25

I’m getting a complete reconstruction using side views and top views:
snapshot00


#26

I will check it again tomorrow.

I was looking at the terminal output…after opensfm ‘reconstruct’ command was completed, it displayed the number of partial reconstructions as 2. However it took only one at the end for further processing in pipeline…

I might be wrong, will test it again.

Thank you so much for your support and coordination with opensfm contributors :slight_smile:


#27

It could be it’s still getting two – I was judging it by the quality of the final point cloud. If there’s a second still, it could be for other reasons.

Cheers!


#28

@smathermather-cm
I tested the new changes on the dataset that I shared earlier.
Yes, it improved the point cloud quality because it was able to use all the images and created only one partial reconstruction from them.

However, when I tested on other similar datasets - it was not able to use all images at the same time and created two partial reconstructions.
I also tried them on opensfm built separately from latest cloned through github- there it created only one partial reconstruction.

**

Is it possible to know why ODM creates two partial reconstructions from the group of images covering same area but through different flight paths?

**
But, the point cloud quality has improved from earlier - now I need not separate images from different flight paths.

Thanks a lot!


#29

Sounds like there’s some benefit to pulling in the latest OpenSfM changes then. I was being conservative I’m only pulling in the patch for matching.


#30

Yes, I am trying to compare latest OpenSfm code with the one that is integrated with present ODM - to understand why the former was creating only one partial reconstruction where as ODM creates two.

But the dense point cloud that comes from Opensfm is very noisy and does not look accurate.

Thank you.