SHARC: Reference point driven Spherical Harmonic Representation for Complex Shapes

This repository contains the implementation of SHARC (Reference point driven Spherical Harmonic Representation for Complex Shapes), a method for efficient 3D shape representation and reconstruction using Spherical Harmonics (SH) centered at optimally selected internal reference points.

Installation

1. Create Environment

   conda create -n sharc python=3.10
   conda activate sharc

2. Install Dependencies Install the required packages from requirements.txt:

   pip install -r requirements.txt

Usage

1. Mesh to SHARC representation

Convert a 3D mesh (e.g., .off, .obj, .ply) into the SHARC representation.

python mesh_to_sharc.py \
    --input_mesh assets/demo/dragon.off \
    --output_dir assets/demo/output/ \
    --lfit 64 \
    --max_reference_points 1000

2. Reconstruction (SHARC to Mesh)

To reconstruct the mesh from the saved coefficients:

python sharc_to_mesh.py \
    --model_path assets/demo/output/dragon/coeffs/sh_coeffs.npz \
    --output_dir assets/demo/output/dragon/reconstruction \
    --mesh