Alpine - a pytorch library for implicit neural representations.#

Alpine is a flexible, user-friendly, and distributed PyTorch library for Implicit Neural Representations. Alpine provides a systematic way to build, train, and visualize neural fields or neural representations across a wide variety of scientific and medical data. Alpine is user-frieldly, scalable, and easily integrates with exiting PyTorch workflows.

Getting started#

import alpine
import torch
import skimage.io, skimage.transform, skimage.data

# load data signal
H, W = 256, 256
data = skimage.transform.resize(skimage.data.astronaut(), (H,W))

# create a pytorch tensor for data signal
data_tensor = torch.from_numpy(data).float().cuda()

# initialize coordinate grid for fitting the signal
coords = alpine.utils.coords_spatial_nd(H,W)[None,...].cuda()

# initialize neural representation
model = alpine.models.Siren(in_features=2, out_features=3).cuda()

# fit signal
result = model.fit_signal(input = coords, signal = data_tensor, n_iters=1000)

For more instructions on how to install alpine please see our installation instructions. Alpine also provides a wide variety of examples detailing various features and functionalities of alpine.

Key Features#

Flexible

Alpine provides modular interfaces for prototyping and testing neural representations. Alpine also offers highly reconfigurable training workflows that allow for custom forward model via closures and loss functions.

Distributed

We use PyTorch lightning to scale Alpine across multiple GPUs. Alpine offers seamless integration with Pytorch lightning using its custom lightining trainer.

Interpretability

Alpine provides a rich set of visualization tools ( and many more to be added) that allow you to peer deep into the features learned by the INR.

Extensible

Written in PyTorch, Alpine can be seamlessly integrated with other PyTorch libraries and models.

Important

Alpine is work in progress. We appreciate any constructive community feedback and support. We invite all the researchers across all disciplines to explore, and suggest any features you find particularly useful. Our goal is to make Alpine the go-to place for scientific computing using INRs!

Citing Alpine#

If you find Alpine useful for your work, please consider citing us

@software{vyas_alpine_2025
    author={ Vyas, Kushal and Saragadam, Vishwanath and Veeraraghavan, Ashok and Balakrishnan, Guha},
    title = {Alpine - A PyTorch Library for Implicit Neural Representations},
    year={2025},
    url={https://github.com/kushalvyas/alpine}
}

@misc{vyas_alpine_2025,
    author = {Vyas, Kushal and Saragadam, Vishwanath and Veeraraghavan, Ashok and Balakrishnan, Guha},
    title = {Alpine - A PyTorch Library for Implicit Neural Representations},
    booktitle={CVPR Workshop Neural Fields Beyond Conventional Cameras}
    year={2025}
}

Quick links: