search

alicevision / Meshroom

3D Reconstruction Software
http://alicevision.org
Other
10.81k stars 1.06k forks source link

[question]How to use different set of image for texturing? #1559

Closed aeoleader closed 2 years ago

aeoleader commented 2 years ago

I am trying to use downscaled image sets for sfm and mesh reconstruction and use the original images for texturing. Since the feature extraction node doesn't offer image downscale options I need to preprocess the input images first then perform feature extraction and the following calculations. However, to ensure the final quality of the output mesh I want to calculate texture based on the original image sets. Is there a way to do so?

I checked the previous issues and the closest matching problem is #763 but it is using the previous version of meshroom, the new version we need to provide Dense SfMData to the texturing node.

natowi commented 2 years ago

I am trying to use downscaled image sets

You can, but why?

Use the method described here: https://github.com/alicevision/meshroom/wiki/Projected-Light-Patterns

aeoleader commented 2 years ago

I am trying to use downscaled image sets

You can, but why?

Use the method described here: https://github.com/alicevision/meshroom/wiki/Projected-Light-Patterns

Because my image is too large to use the popsift method. It always fails with not enough cuda memory error, but I still want the best texture quality. I followed the method you mentioned but get the dimension mismatch error. Below is my workflow and the error message. image image

natowi commented 2 years ago

Popsift hast a limit on the image size https://github.com/alicevision/popsift/issues/77 I don´t know how to solve your dimension mismatch problem for now. Maybe dsp-sift works with the full image?.

aeoleader commented 2 years ago

Popsift hast a limit on the image size alicevision/popsift#77 I don´t know how to solve your dimension mismatch problem for now. Maybe dsp-sift works with the full image?.

Thanks for your reply!

Would switching to a GPU with larger VRAM help?

aeoleader commented 2 years ago

BTW, I think maybe a feature request can be made regarding adding downscale factor to the feature extraction node, because sometimes it is inefficient to use the fullsize images to perform tracking, and the fullscale images are usually required at the reconstruction stage in practice.

stale[bot] commented 2 years ago

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

stale[bot] commented 2 years ago

This issue is closed due to inactivity. Feel free to re-open if new information is available.