Spherical INR

Spherical INR logo

A PyTorch library for building Implicit Neural Representations on spherical and 3D domains.

Features:

  • Positional Encodings: Herglotz, Fourier, Spherical & Solid Harmonics

  • INR Wrappers: INR, HerglotzNet, SirenNet, and spherical variants

  • Transforms: Cartesian ↔ Spherical ↔ Polar

  • MLP backbones: Standard & Sine-activated (SineMLP)

  • Losses & Differentiation utilities

Getting Started

Install from PyPI:

pip install spherical-inr

Or for development:

git clone https://github.com/your-org/spherical-inr.git
cd spherical-inr
pip install -e .

Quick Example

import torch
from spherical_inr import SphericalSirenNet, tp_to_r3

# build a spherical SIREN: L=3 harmonics, two hidden layers of size 64, output dim=1
net = SphericalSirenNet(L=3, mlp_sizes=[64,64], output_dim=1, seed=0)

# sample some θ,ϕ in radians
coords = torch.rand(8,2) * torch.tensor([3.1416, 6.2832])
y = net(coords)  # forward on sphere

API Reference

Core Modules