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

Comparison before and after compensation PR #1

Open
jb-ye opened this issue Feb 1, 2024 · 8 comments
Open

Comparison before and after compensation PR #1

jb-ye opened this issue Feb 1, 2024 · 8 comments

Comments

@jb-ye
Copy link
Owner

jb-ye commented Feb 1, 2024

Original behavior. Best quality was rendered at 1000 pix, artifacts observed when rendering at 500 or 2000.

PR_before.mp4

New behavior after rendering

PR_after.mp4
@jb-ye
Copy link
Owner Author

jb-ye commented Feb 1, 2024

Applying the change, we no longer observe "aliasing" effect.

bike.mp4

@jb-ye
Copy link
Owner Author

jb-ye commented Feb 2, 2024

More direct comparison before and after

output.mp4

@ichsan2895
Copy link

What a great improvement

@arlo-ml
Copy link

arlo-ml commented Feb 9, 2024

Hi, this is a really interesting solution
I only noticed a few artifacts on the spokes and the wheels, as if they were mixed together:

Screenshot 2024-02-09 130201
Screenshot 2024-02-09 132919

@ichsan2895
Copy link

I tested on another dataset, so far only 4 tested (kitchen, truck, playroom, and desolation). Where 3 of them (truck, playroom, and desolation) I see a little bit improvement.

Kitchen dataset with splatfacto-v101:
0000_Classic-NewRasterizer
0005_Classic-NewRasterizer

Kitchen dataset with --pipeline.model.rasterize-mode 'standard':
0000_Standard-NewRasterizer
0005_Standard-NewRasterizer

playroom dataset:
Left = Using --pipeline.model.rasterize-mode 'standard'
Right = using Splatfacto v1.0.1, so it don't use this rasterization.

Standar_vs_Classic_NewRasterizer_0004
Standar_vs_Classic_NewRasterizer_0003
Standar_vs_Classic_NewRasterizer_0002
Standar_vs_Classic_NewRasterizer_0001

@jb-ye
Copy link
Owner Author

jb-ye commented Feb 9, 2024

@ichsan2895 which viewer are you using? Currently only nerfstudio viewer is supported with this feature. Other viewer doesn’t display the results correctly.

@jb-ye
Copy link
Owner Author

jb-ye commented Feb 9, 2024

Hi, this is a really interesting solution I only noticed a few artifacts on the spokes and the wheels, as if they were mixed together:

Screenshot 2024-02-09 130201 Screenshot 2024-02-09 132919

This is possible if there is no close view of the object in the input images.

@ichsan2895
Copy link

ichsan2895 commented Feb 9, 2024

@ichsan2895 which viewer are you using? Currently only nerfstudio viewer is supported with this feature. Other viewer doesn’t display the results correctly.

kitchen dataset: it was a frame from video which come from ns-render result. So, it does not screenshot the 3DGS Web viewer.

playroom dataset: it was the screenshot from mkellogg 3D WebGL Gaussian Splat.

jb-ye pushed a commit that referenced this issue Aug 21, 2024
* Integrate 3dgs-mcmc into gsplat. (#1)

* Initialize from random point cloud (nerfstudio-project#2)

* Clean up simple trainer changes. (nerfstudio-project#3)

* formatting

* Create dedicated mcmc train script. (nerfstudio-project#4)

* minor

* scene scale bug

* scheduler

* PR comments wip (nerfstudio-project#5)

* Some cuda changes (nerfstudio-project#6)

* data to data_ptr

* use scale_rot_to_cov3d from cuda_legacy

* quat_scale_to_covar_preci

* removed _sample_new_gaussians

* remove old prefix

* zeros_like

* empty_like

* cuda stream

* min opacity
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

3 participants