Export to MVE format?

[Dn Jiang]

Hi Chris,

Is there any plan to export output into MVE format? PMVS is GPL and is not scalable for large data set. MVE is BSD licensed and supports large data set, plus texturing. I can see many benefit of MVE if it is what it claims (I am starting to try it).

Thanks,

Dan

[Benjamin Nuernberger]

Hi Dan,

I've successfully exported Theia reconstructions into bundler files which I then fed into MVE. I followed this process which has worked so far for me:

1. mkdir /path/to/mve/datasets/
2. mkdir /path/to/mve/datasets/my_new_dataset/
3. In the following instructions, <scene-dir> is /path/to/mve/datasets/my_new_dataset.
4. Make a directory that will contain a "bundler" reconstruction: mkdir <scene-dir>/bundler_dir/.

5. Convert the Theia reconstruction to a "bundler" format via:

   ./convert_theia_reconstruction_to_bundler_file --reconstruction_file=/path/to/reconstruction --bundle_file=<scene-dir>/bundler_dir/bundle/bundle.out --lists_file=<scene-dir>/bundler_dir/list.txt

6. Copy your images into <scene-dir>/bundler_dir

7. cd /path/to/mve/apps/makescene
8. makescene -i <scene-dir>/images <scene-dir>

And then simply follow the rest of the MVE instructions on its website.

Thanks,

Ben

--
You received this message because you are subscribed to the Google Groups "Theia Vision Library" group.
To unsubscribe from this group and stop receiving emails from it, send an email to theia-vision-lib...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Dn Jiang]

Thanks Ben, will give it s shot.

Now another question for you: what is your experience with MVE in terms of stability and scalability?

Dan

To unsubscribe from this group and stop receiving emails from it, send an email to theia-vision-library+unsub...@googlegroups.com.

[Benjamin Nuernberger]

I've only made a couple of reconstructions with MVE, so I can't really say anything definitively. I think I ran one reconstruction a couple of times and it seemed fairly stable both times. Hard to say anything about scalability.

Dan

To unsubscribe from this group and stop receiving emails from it, send an email to theia-vision-lib...@googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "Theia Vision Library" group.

To unsubscribe from this group and stop receiving emails from it, send an email to theia-vision-lib...@googlegroups.com.

[Benjamin Nuernberger]

Sorry, I think the steps I sent you had some typos or missed some things; here's an updated set of steps:

1. mkdir /path/to/mve/datasets/
2. mkdir /path/to/mve/datasets/my_new_dataset/
3. In the following instructions, <scene-dir> is /path/to/mve/datasets/my_new_dataset.

1. Make a directory that will contain a "bundler" reconstruction: mkdir <scene-dir>/bundler_dir/ and mkdir <scene-dir>/bundler_dir/bundle/.

1. Convert the Theia reconstruction to a "bundler" format via:

   ./convert_theia_reconstruction_to_bundler_file --reconstruction_file=/path/to/reconstruction --bundle_file=<scene-dir>/bundler_dir/bundle/bundle.out --lists_file=<scene-dir>/bundler_dir/list.txt

1. Copy your images into <scene-dir>/images/

2. cd /path/to/mve/apps/makescene
3. ./makescene -i <scene-dir>/images <scene-dir>

Let me know if this works or needs to be updated for future reference.

Thanks,
Ben

[Dn Jiang]

Hi Ben,

Sorry for the late reply. But I have been trying to run MVE alone (w/o using Theia for sfm) with my own sample data set. Unfortunately, I have been getting crashes (I am running MVE on Windows). I'll post the result of the your above procedure once I get a chance to try it (i.e., after getting my sample data working with MVE first).

Thanks,

Dan

Dan

To unsubscribe from this group and stop receiving emails from it, send an email to theia-vision-library+unsub...@googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "Theia Vision Library" group.

To unsubscribe from this group and stop receiving emails from it, send an email to theia-vision-library+unsub...@googlegroups.com.

[Chris Sweeney]

Dan,

You may want to reach out to the MVE authors on github. They are very responsive and helpful, and I'm sure they could help you get MVE to work if you are having issues.

Chris

Dan

To unsubscribe from this group and stop receiving emails from it, send an email to theia-vision-lib...@googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "Theia Vision Library" group.

To unsubscribe from this group and stop receiving emails from it, send an email to theia-vision-lib...@googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "Theia Vision Library" group.

To unsubscribe from this group and stop receiving emails from it, send an email to theia-vision-lib...@googlegroups.com.

[forsale...@gmail.com]

On Thursday, June 2, 2016 at 2:41:40 PM UTC-4, Benjamin Nuernberger wrote:
> Sorry, I think the steps I sent you had some typos or missed some things; here's an updated set of steps:
>

> mkdir /path/to/mve/datasets/mkdir /path/to/mve/datasets/my_new_dataset/In the following instructions, <scene-dir> is /path/to/mve/datasets/my_new_dataset.Make a directory that will contain a "bundler" reconstruction: mkdir <scene-dir>/bundler_dir/ and mkdir <scene-dir>/bundler_dir/bundle/.

> Convert the Theia reconstruction to a "bundler" format via:
> ./convert_theia_reconstruction_to_bundler_file --reconstruction_file=/path/to/reconstruction --bundle_file=<scene-dir>/bundler_dir/bundle/bundle.out --lists_file=<scene-dir>/bundler_dir/list.txt

> Copy your images into <scene-dir>/images/cd /path/to/mve/apps/makescene./makescene -i <scene-dir>/images <scene-dir>Let me know if this works or needs to be updated for future reference.

>
> Thanks,
> Ben
>
>
> On Mon, May 30, 2016 at 7:38 PM, Benjamin Nuernberger <bnuern...@cs.ucsb.edu> wrote:
>
>
> I've only made a couple of reconstructions with MVE, so I can't really say anything definitively. I think I ran one reconstruction a couple of times and it seemed fairly stable both times. Hard to say anything about scalability.
>
>
>
>
>
> On Mon, May 30, 2016 at 3:08 AM, Dn Jiang <dn.j...@gmail.com> wrote:
>
> Thanks Ben, will give it s shot.
>
>
> Now another question for you: what is your experience with MVE in terms of stability and scalability?
>
>
> Dan
>
>
> On Sunday, May 29, 2016 at 9:50:44 PM UTC-4, Benjamin Nuernberger wrote:
> Hi Dan,
>
>
> I've successfully exported Theia reconstructions into bundler files which I then fed into MVE. I followed this process which has worked so far for me:
>
>

> mkdir /path/to/mve/datasets/mkdir /path/to/mve/datasets/my_new_dataset/In the following instructions, <scene-dir> is /path/to/mve/datasets/my_new_dataset.Make a directory that will contain a "bundler" reconstruction: mkdir <scene-dir>/bundler_dir/.

> Convert the Theia reconstruction to a "bundler" format via:
> ./convert_theia_reconstruction_to_bundler_file --reconstruction_file=/path/to/reconstruction --bundle_file=<scene-dir>/bundler_dir/bundle/bundle.out --lists_file=<scene-dir>/bundler_dir/list.txt

> Copy your images into <scene-dir>/bundler_dircd /path/to/mve/apps/makescenemakescene -i <scene-dir>/images <scene-dir>And then simply follow the rest of the MVE instructions on its website.

>
>
> Thanks,
> Ben
>
>
> On Sun, May 29, 2016 at 6:06 AM, Dn Jiang <dn.j...@gmail.com> wrote:
>
> Hi Chris,
>
>
> Is there any plan to export output into MVE format? PMVS is GPL and is not scalable for large data set. MVE is BSD licensed and supports large data set, plus texturing. I can see many benefit of MVE if it is what it claims (I am starting to try it).
>
>
> Thanks,
> Dan
>
>
>
>
> --
>
> You received this message because you are subscribed to the Google Groups "Theia Vision Library" group.
>
> To unsubscribe from this group and stop receiving emails from it, send an email to theia-vision-lib...@googlegroups.com.
>
> For more options, visit https://groups.google.com/d/optout.
>
>
>
>
>
>
>
>
>
>
> --
>
> You received this message because you are subscribed to the Google Groups "Theia Vision Library" group.
>
> To unsubscribe from this group and stop receiving emails from it, send an email to theia-vision-lib...@googlegroups.com.
>
> For more options, visit https://groups.google.com/d/optout.

Hi Ben,

I have managed to follow your procedure. The only step I need to change is Step 6. I had to copy my input images into <scene-dir>/bundler_dir/ instead of <scene-dir>/images/.

But I still have problem using the Theia sfm output with MVE.

I am able to run makescene and then dmrecon to create non-empty *.mvei depth maps, but then scene2pset always produces 0 points no matter what scale I use. The pset-L0.ply or pset-L2.ply file created by scene2pset.exe contains header only.

With the same data set, I am able to visualize the theia's sfm output and then am able to import into pmvs to see dense point cloud. But I have not been able to create any point cloud with MVE's scene2pset.exe.

I am not sure if this is due to Theia's convert_theia_reconstruction_to_bundler.exe or
MVE's scene2pset.exe. As a result, I do not know whom to report the issue to or ask for help from.

Thanks,
Dan

[Chris Sweeney]

I am happy to help out with Theia-related bugs in any form, but in this case it seems that it would be easier to debug from MVE's side (even if the error is caused by Theia) since the symptoms are only showing in MVE's programs.

Is there another way I could offer help? I am a bit unsure on the best way to advise.

Chris

[Dn Jiang]

Hi Chris,

I just created a detailed ticket with MVE:

https://github.com/simonfuhrmann/mve/issues/239

Finger crossed.

Thanks,

Dan

[Dn Jiang]

Hi Chris,

The MVE developer replied and closed the ticket with the following comment:

https://github.com/simonfuhrmann/mve/issues/239

"That's because your depth maps don't contain any reconstructed pixels. Probably because you don't have enough views for depth map reconstruction"

The thing is: I exported the same sfm output to PMVS which does reconstruct a dense cloud. My guess is that the issue could be caused by "convert_theia_reconstruction_to_bundler_file.exe" ?

Wonder if you could investigate this issue (it is out of my current knowledge) and I can send you my input image set upon confirmation.

Thanks,

Dan

[Chris Sweeney]

What happens when you convert your reconstruction to a bundler file, then convert that bundler file back to a Theia file. Can you confirm the two reconstructions are equivalent?

Chris

[Chris Sweeney]

Note that you can use the compare_reconstructions.cc file to evaluate any differences between the reconstructions. That will compare views that contain the same name in both reconstructions (which should be all views for your case).

[Dn Jiang]

I did the following:

.\VS2013\bin\Release\convert_theia_reconstruction_to_bundler_file --reconstruction_file=.\Output\ET\Reconstruction_0-0 --bundle_file=.\Output\ET\MVE\bundler_dir\bundle\bundle.out --lists_file=.\Output\ET\MVE\bundler_dir\list.txt

.\VS2013\bin\Release\convert_bundle_file.exe --bundle_file=.\Output\ET\MVE\bundler_dir\bundle\bundle.out --lists_file=.\Output\ET\MVE\bundler_dir\list.txt --output_reconstruction_file=.\Output\ET\Reconstruction_0-0-from-bundle

.\VS2013\bin\Release\compare_reconstructions.exe -reconstruction_to_align=.\Output\ET\Reconstruction_0-0-from-bundle --reference_reconstruction=.\Output\ET\Reconstruction_0-0

The result of compare_reconstruction seems to confirm the two reconstructions are the same:

Number of cameras:

        Reconstruction 1: 4

        Reconstruction 2: 4

        Number of Common cameras: 4Number of 3d points:

        Reconstruction 1: 637

        Reconstruction 2: 637Rotation difference when aligning orientations:

Mean = 1.567581

Median = 1.567581

Histogram:

[1.000000 - 2.000000) = 4

[2.000000 - 5.000000) = 0

[5.000000 - 10.000000) = 0

[10.000000 - 15.000000) = 0

[15.000000 - 20.000000) = 0

[20.000000 - 45.000000) = 0

Pose error:

Rotation Error:

Mean = 0.000000

Median = 0.000000

Histogram:

< 1.000000 = 4

[1.000000 - 2.000000) = 0

[2.000000 - 5.000000) = 0

[5.000000 - 10.000000) = 0

[10.000000 - 15.000000) = 0

[15.000000 - 20.000000) = 0

[20.000000 - 45.000000) = 0

Position Error:

Mean = 0.000000

Median = 0.000000

Histogram:

< 1.000000 = 4

[1.000000 - 5.000000) = 0

[5.000000 - 10.000000) = 0

[10.000000 - 50.000000) = 0

[50.000000 - 100.000000) = 0

[100.000000 - 1000.000000) = 0

Focal length errors:

Mean = 0.000000

Median = 0.000000

Histogram:

< 0.010000 = 4

[0.010000 - 0.050000) = 0

[0.050000 - 0.200000) = 0

[0.200000 - 0.500000) = 0

[0.500000 - 1.000000) = 0

[1.000000 - 10.000000) = 0

[10.000000 - 100.000000) = 0

Any further hint?

Thanks,

Dan

To unsubscribe from this group and stop receiving emails from it, send an email to theia-vision-library+unsub...@googlegroups.com.

[Dn Jiang]

I ran the same data set using MVE's own SFM module, got the same empty point set from the same input images. So the issue is not with Theia. Already raised a bug with MVE.

-Thanks for all the help.