search

RaduAlexandru / permuto_sdf

Code for our CVPR'23 paper - "PermutoSDF: Fast Multi-View Reconstruction with Implicit Surfaces using Permutohedral Lattices"
https://radualexandru.github.io/permuto_sdf/
MIT License
435 stars 26 forks source link

Head reconstruction #33

Open Zvyozdo4ka opened 1 year ago

Zvyozdo4ka commented 1 year ago

Do you know which tool they employed to get head without background?

Is it necessary to preprocess images and get rid of background?

What is resolution requirement to images?

It is possible to process head with hair and opened eye?

Why do we need imgs_train and imgs_test?

Do you have sample data which you used for processing this model? image

xuduo18311199384 commented 1 year ago

First, remove background is not necessary. Second, if you want to remove background, can use rembg. Third, any image resolution can be used. Of course, the higher the better. Fourth, it can process full head. Final, author already offered custom dataset which is include a full head.

Zvyozdo4ka commented 1 year ago

I appreciate your explanation. Rembg worked well for me.

I ran full head, but could not extract mesh. How did you do this?

RaduAlexandru commented 11 months ago

Hi @Zvyozdo4ka,

You can extract meshes from either sdf model and density models using the functions from this file:

Concretely the function extract_mesh_from_sdf_model can be used on a checkpointed model to extract a mesh from an sdf.