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.

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

Obtaining a more crisp point cloud #1034

Open
arv8315 opened this issue Jan 1, 2024 · 3 comments
Open

Obtaining a more crisp point cloud #1034

arv8315 opened this issue Jan 1, 2024 · 3 comments

Comments

@arv8315
Copy link

arv8315 commented Jan 1, 2024

I am having difficulty with the SFM. It is always generating a fuzzy looking point cloud across multiple datasets. I have attached one example.
3D_point_cloud_buddha

It is based on an orbit shot for a sequence of images for which one image looks as follows:
output_0002_bk_rmvd_with_prompt_1

I am not able to easily find information to make this point cloud come out looking nicer. I have tried increasing the number of frames and the camera has relatively high resolution. I can clean up any motion blur a bit more with some of the shots. However, when trying on other datasets where there is a significant number of frames and no motion blur, this fuzzy point cloud issue persists. Can someone suggest a way to fix this?

There are too many parameters in this config file: https://opensfm.org/docs/_modules/opensfm/config.html. I am doing some checking but not totally sure which adjustment is going to make the difference yet.

@arv8315
Copy link
Author

arv8315 commented Jan 2, 2024

module 'cv2' has no attribute 'xfeatures2d' is the error when trying SURF in place of HAHOG. There might be a fix but have not implemented it and tested that. Using SIFT results in .ply with no data in it! HAHOG takes a shorter time to parse compared to SIFT! Otherwise tried to increase the 'depthmap_min_consistent_views' to a number larger than 3 but it resulted in the 'feature' parts of the point cloud being gouged which is not desirable. Trying to change bundle_outlier_filtering_type from FIXED to AUTO does not make much difference. I have to look further into the source to understand the difference. I tried to change triangulation_type to ROBUST but it did not make much difference.

In contrast, when I tried to use colmap, I got a better sparse point cloud without as many distortions. To get to a Dense PC, it requires a GPU, so I will try that when I get one sorted out shortly. Meanwhile, it would be good if someone has a suggestion to overcome the OpenSFM limitation with generating noisy PCs. I might still be overlooking changing the right setting to improve the outcome.

@arv8315
Copy link
Author

arv8315 commented Jan 2, 2024

I got this as the mesh btw:
image

The facial features are not very clear and all the detail is fuzzy.

@arv8315
Copy link
Author

arv8315 commented Jan 2, 2024

Akaze works but same issue with noisy Point cloud. Does it come down to issue with incomplete EXIF camera parameters? Going to see if that makes a difference.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

1 participant