---
id: pytorch-wavelets
version: "1.3.0"
license: Free To Use
license_treatment: unclear
maintenance: abandoned
---

# pytorch-wavelets — A port of the DTCWT toolbox to run on pytorch

License: unclear · Maintenance: abandoned · Downloads: 198.2K/mo

## What it is and what it does

pytorch-wavelets implements wavelet transforms optimized for PyTorch tensors, letting you compute 2D discrete wavelet transforms (DWT) and dual-tree complex wavelet transforms (DTCWT) on batches of images with full gradient support. It also includes a scattering network layer based on DTCWT. The package uses standard PyTorch NCHW format and supports both CPU and GPU execution.

The library provides forward and inverse transforms for both DWT and DTCWT, plus specialized scattering layers (ScatLayer, ScatLayerj2) for feature extraction. Version 1.3.0 added 1D wavelet support. However, the project is no longer maintained—the last commit was in August 2023—so it may not work with current versions of PyTorch or NumPy without modification.

Use it for:

- Extract multiscale wavelet features from image batches for machine learning pipelines.
- Implement learnable wavelet-based layers in neural networks with backpropagation through transforms.
- Compute scattering network representations for image classification or analysis on GPU.
- Perform 1D wavelet decomposition on time-series or signal data within PyTorch models.
- Prototype research using dual-tree complex wavelets without leaving the PyTorch ecosystem.

## Worth the install?

AI-flagged interpretation of the facts above — verify before relying.

Provides 2D discrete wavelet and dual-tree complex wavelet transforms (DWT and DTCWT) with their inverses in PyTorch, supporting gradient backpropagation and batched multichannel image processing.

Yes, if you need wavelet transforms in PyTorch and can tolerate an abandoned package. Install friction is low and there are no known vulnerabilities. However, verify that the 'Free To Use' license suits your use case, test compatibility with your PyTorch and NumPy versions, and be prepared to maintain or fork the code if bugs emerge—no updates are coming from upstream.

## Install

pip install pytorch-wavelets
uv add pytorch-wavelets
poetry add pytorch-wavelets

## Installing pytorch-wavelets

Before you install: Low install friction with a pure Python wheel. Maintenance is abandoned—last release was 1.3.0 on 2023-04-13, with no commits since 2023-08-02—so expect no bug fixes or updates.

License in practice: License is marked 'Free To Use' with unclear treatment and no SPDX identifier. Consult the repository's ORIGINAL_README.txt and license terms before use in proprietary or commercial contexts.

Quickstart:

pip install pytorch-wavelets

import torch
from pytorch_wavelets import DWTForward, DWTInverse

xfm = DWTForward(J=3, wave='db3', mode='zero')
X = torch.randn(10, 5, 64, 64)
Yl, Yh = xfm(X)
print(Yl.shape)  # torch.Size([10, 5, 12, 12])

ifm = DWTInverse(wave='db3', mode='zero')
Y = ifm((Yl, Yh))

Verify before relying:

- Whether the 'Free To Use' license permits commercial or proprietary use without restriction.
- Current compatibility with recent PyTorch and NumPy versions given the package's abandonment.
- Whether GPU/CUDA support requires additional system dependencies beyond PyTorch.

## Package facts

- License: Free To Use (unclear)
- Python support: unspecified
- Install friction: low
- Maintenance: abandoned
- Downloads: 198.2K/month (top 15,000 on PyPI)
- Known vulnerabilities: none known

## Tags

wavelet transform pytorch, discrete wavelet transform gpu, DTCWT complex wavelet, image wavelet decomposition, differentiable wavelet layers, scattering network pytorch, 1D wavelet transform, wavelet-analysis, signal-processing, gpu-accelerated

[View on SkillFed](https://skillfed.io/packages/pytorch-wavelets) · [View on PyPI](https://pypi.org/project/pytorch-wavelets/)