https://github.com/chris-official/pytorchgaf

PyTorch accelerated GAF transform
https://github.com/chris-official/pytorchgaf

cuda gpu gramian-angular-fields image-analysis python pytorch time-series

Last synced: about 1 month ago
JSON representation

PyTorch accelerated GAF transform

• Host: GitHub
• URL: https://github.com/chris-official/pytorchgaf
• Owner: chris-official
• License: mit
• Created: 2025-04-03T13:55:43.000Z (about 1 year ago)
• Default Branch: main
• Last Pushed: 2025-04-05T13:07:46.000Z (about 1 year ago)
• Last Synced: 2025-04-09T16:20:01.118Z (about 1 year ago)
• Topics: cuda, gpu, gramian-angular-fields, image-analysis, python, pytorch, time-series
• Language: Python
• Homepage:
• Size: 101 KB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          
# PyTorchGAF: PyTorch-accelerated implementation of the Gramian Angular Field (GAF)

The provided GAF implementation uses pure PyTorch. It provides the following benefits and features:

- Supports batched data.

- Supports multivariate time series data.

- Handles necessary scaling before the GAF transformation.

- Allows to use the GAF transform directly as a neural network layer.

- Allows directly transforming the data on the target device (CPU or GPU).

- Uses efficient vectorized Einstein Summation Notation to compute the outer products to avoid using loops.

## Optimizations

Our implementation is based on the GAF transform from pyts. However, as Figure 1 shows, the GAF transform significantly benefits from GPU acceleration, achieving speedups of almost 90x for larger batch sizes.

![Performance Comparison](performance_comparison.png)

*Figure 1: Execution time comparison between CPU-based pyts implementation and our implementation on CPU and GPU.*

## Usage/Examples

```python

import torch

gaf = GAFTransform(method="summation")

inputs = torch.randn(32, 8, 40, device="cuda")  # (N, C, L)

output = gaf(inputs)  # (N, C, L, L)

```

## Acknowledgements

 - [pyts GitHub](https://github.com/johannfaouzi/pyts)

 - [pyts GAF](https://github.com/johannfaouzi/pyts/blob/main/pyts/image/gaf.py)