Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Cannot reproduce results in the paper #1

Closed
lxxue opened this issue Dec 1, 2020 · 4 comments
Closed

Cannot reproduce results in the paper #1

lxxue opened this issue Dec 1, 2020 · 4 comments

Comments

@lxxue
Copy link

lxxue commented Dec 1, 2020

Hi Qingan,

Thanks for the interesting work and sharing the code. I tried run it with the newest bundler_sfm on the Temple of Heaven dataset and it gives me such results:

bundler_TOH

It seems that the result is not correct as in the paper the cameras are all around the scene. Therefore, half of the scene is not reconstructed correctly. I wonder if you have tried this bundler version on the TOH dataset and got similar results in the paper?

I tried with the bundler_sfm with its example dataset kermit and the reconstruction looks similar to the reference, so I guess the bundler part is correct.

Also a side question, how do you usually view the reconstruction visually and check if the result is better or not? Is it through the reprojection error and number of tracks? I find the rendered images in the paper are really nice and clear.
@llcc343
Copy link

llcc343 commented Dec 4, 2020

Same problems here. Even with finetune the parameters the author recommended, the results are still very bad. I think the results in their paper are fake results. Very sad:(

@lxxue
Copy link
Author

lxxue commented Dec 4, 2020

Actually, I just got a nice result:
image
The difference is that now I use the version of Bundler in the repo, with a modification in line 9 of src/Makefile:

OTHERFLAGS=-std=gnu++0x -fPIC

so that it compiles successfully on ubuntu 18.04. (also need to replace one file in lib/sba-1.5 as instructed here: snavely/bundler_sfm#74 (comment)).

I will check the reason why it didn't work in the first place later.

@llcc343
Copy link

llcc343 commented Dec 5, 2020

Well, I did not use sba for optimization, and use ceres solver in bundler sfm. But the ambiguation is not eliminated. Have you ever test on other ambiguous datasets?

@lxxue
Copy link
Author

lxxue commented Dec 8, 2020

Yeah, it works on the Alexander Nevsky Cathedral dataset. But there are also quite some failure cases.

@lxxue lxxue closed this as completed Dec 9, 2020
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
2 participants
@llcc343 @lxxue