smal_fitter

This library of scripts provides Mesh-to-Mesh fitting and model generation of a modified version of the SMAL model, introduced in the paper '3D Menagerie: Modeling the 3D shape and pose of animals' (Zuffi et al, 2016).

The scripts are intended to produce a new version of the model from a set of input .obj files, and is capable of

Optimising the SMBLD model to .obj models, and then applying vertex deformations for a smooth fit.
Fitting this model to animation sequences, provided as .obj files.
Using these fits to produce a new shape space for the models generated.
Visualising shape spaces and animation sequences.

Mesh fitting

Mesh fitting allows for the modification of paramaterised meshes, or vertex deformations, to fit an input mesh to the .obj files provided.

As Rigid As Possible (ARAP) regularisation can be provided for mesh fitting. This requires the pytorch-arap module to also be installed (see setup).

Setup

To install, clone the repository with pytorch_arap submodule, using

git clone --recurse-submodules https://github.com/OllieBoyne/smal-fitter

OPTIONAL: For improved ARAP Speed, install torch_batch_svd, from https://github.com/KinglittleQ/torch-batch-svd. If this is not used, torch.svd will be used.

Dataset

This module was trained using the Unity Dog Big Pack https://assetstore.unity.com/packages/3d/characters/dog-big-pack-105660, but will work on any set of .obj files.