[](https://opensource.org/licenses/MIT)
# PyInterpX - Higher Order Interpolation in 3D for Torch
<p align="center">
<img src="https://github.com/ThomasHelfer/PyInterpX/blob/main/img/logo_cropped.png" alt="no alignment" width="25%" height="auto"/>
</p>
## Overview
PyInterpX is a compact library designed for advanced 3D interpolation using higher order polynomial bases, which is not currently supported by PyTorch's `torch.nn.functional.interpolate()` method. This enhancement allows for more precise and customized interpolation processes in 3D spaces, catering to specialized applications requiring beyond-linear data manipulation.
## Quick Start
To get started with PyInterpX:
1. Install the library using pip:
```bash
pip install pyinterpx
```
2. Import `interp` from PyInterpX and PyTorch in your script or notebook:
```python
from pyinterpx.Interpolation import interp
import torch
```
3. Utilize the interpolation function with a 6x6x6 kernel, polynomials up to the third power, and 25 channels:
```python
points, power, channels = 6, 3, 25
Interp = interp(points, power, channels)
x = torch.rand(2, 25, 10, 10, 10)
Interp(x)
```
## Key Features
- **Fast**: Optimized for high performance across any device.
- **CPU and GPU Compatible**: Functions seamlessly on both CPU and GPU environments.
```python
points, power, channels = 6, 3, 25
# Running on GPU for even faster computations
interp = interp(points, power, channels, device="cuda:0")
```
- **Precise**: Supports various data types for precise computation.
```python
points, power, channels = 6, 3, 25
# Using double for more precision
interp = interp(points, power, channels, dtype=torch.double)
```
- **Integrated with PyTorch**: Easily integrates within the PyTorch ecosystem.
```python
# A simple model which uses interpolation at some layer
class Model(torch.nn.Module):
def __init__(self):
super(Model, self).__init__()
points, power, channels = 6, 3, 25
# Setting up interpolation
self.interpolation = interp(points, power, channels)
self.convs = torch.nn.Sequential(
torch.nn.Conv3d(25, 64, kernel_size=3, padding=1),
torch.nn.ReLU(inplace=True),
)
def forward(self, x):
x = self.convs(x)
x = self.interpolation(x)
return x
```
- Choose simply the **grid alignment** you like.
```python
points, power, channels = 6, 3, 25
interp = interp(points, power, channels, dtype=torch.double,align_corners = False)
```
<p align="center">
<img src="https://github.com/ThomasHelfer/HigherOrderInterpolation3DTorch/blob/main/img/no_align.png" alt="no alignment" width="50%" height="auto"/>
</p>
or if you do not want to have any aligment with the input grid
```python
points, power, channels = 6, 3, 25
interp = interp(points, power, channels, dtype=torch.double,align_corners = True)
```
<p align="center">
<img src="https://github.com/ThomasHelfer/HigherOrderInterpolation3DTorch/blob/main/img/align.png" alt="aligned" width="50%" height="auto"/>
</p>
- Choose the enhacement **factor** you like
```python
factor = 4
points, power, channels = 6, 3, 25
interp = interp(points, power, channels, dtype=torch.double,align_corners = False,factor = factor)
```
<p align="center">
<img src="https://github.com/ThomasHelfer/HigherOrderInterpolation3DTorch/blob/main/img/interpolation_grid_zoomed_factor_4.png" alt="no alignment" width="50%" height="auto"/>
</p>
```python
factor = 16
points, power, channels = 6, 3, 25
interp = interp(points, power, channels, dtype=torch.double,align_corners = False,factor = factor)
```
<p align="center">
<img src="https://github.com/ThomasHelfer/HigherOrderInterpolation3DTorch/blob/main/img/interpolation_grid_zoomed_factor_16.png" alt="no alignment" width="50%" height="auto"/>
</p>
### Prerequisites
Before installing PyInterpX, ensure you meet the following prerequisites:
- Python 3.8 or higher
- pip package manager
## License
PyInterpX is open-sourced under the MIT License. For more details, see the LICENSE file.
## Contact
For inquiries or support, reach out to Thomas Helfer at thomashelfer@live.de.
Raw data
{
"_id": null,
"home_page": "",
"name": "PyInterpX",
"maintainer": "",
"docs_url": null,
"requires_python": ">=3.8",
"maintainer_email": "",
"keywords": "Machine learning,Pytorch,Interpolation,Higher order interpolation,3D interpolation,GPU compatible",
"author": "ThomasHelfer",
"author_email": "thomashelfer@live.de",
"download_url": "https://files.pythonhosted.org/packages/dd/12/e0f3d56fab2c80186fea50b15e7a4b6e9b807a04aedbb5791a8160bfd2c8/PyInterpX-0.2.3.tar.gz",
"platform": null,
"description": "\n[](https://opensource.org/licenses/MIT)\n\n# PyInterpX - Higher Order Interpolation in 3D for Torch\n<p align=\"center\">\n <img src=\"https://github.com/ThomasHelfer/PyInterpX/blob/main/img/logo_cropped.png\" alt=\"no alignment\" width=\"25%\" height=\"auto\"/>\n</p>\n\n## Overview\n\nPyInterpX is a compact library designed for advanced 3D interpolation using higher order polynomial bases, which is not currently supported by PyTorch's `torch.nn.functional.interpolate()` method. This enhancement allows for more precise and customized interpolation processes in 3D spaces, catering to specialized applications requiring beyond-linear data manipulation.\n\n## Quick Start\n\nTo get started with PyInterpX:\n\n1. Install the library using pip:\n\n ```bash\n pip install pyinterpx\n ```\n\n2. Import `interp` from PyInterpX and PyTorch in your script or notebook:\n\n ```python\n from pyinterpx.Interpolation import interp\n import torch\n ```\n\n3. Utilize the interpolation function with a 6x6x6 kernel, polynomials up to the third power, and 25 channels:\n\n ```python\n points, power, channels = 6, 3, 25\n Interp = interp(points, power, channels)\n x = torch.rand(2, 25, 10, 10, 10)\n Interp(x)\n ```\n\n## Key Features\n\n- **Fast**: Optimized for high performance across any device.\n\n![Performance Comparison](https://github.com/ThomasHelfer/HigherOrderInterpolation3DTorch/blob/main/img/Comparison.png \"Performance Comparison\")\n\n- **CPU and GPU Compatible**: Functions seamlessly on both CPU and GPU environments.\n\n ```python\n points, power, channels = 6, 3, 25\n # Running on GPU for even faster computations \n interp = interp(points, power, channels, device=\"cuda:0\")\n ```\n\n- **Precise**: Supports various data types for precise computation.\n\n ```python\n points, power, channels = 6, 3, 25\n # Using double for more precision \n interp = interp(points, power, channels, dtype=torch.double)\n ```\n\n- **Integrated with PyTorch**: Easily integrates within the PyTorch ecosystem.\n\n ```python\n # A simple model which uses interpolation at some layer\n class Model(torch.nn.Module):\n def __init__(self):\n super(Model, self).__init__()\n points, power, channels = 6, 3, 25\n # Setting up interpolation \n self.interpolation = interp(points, power, channels)\n\n self.convs = torch.nn.Sequential(\n torch.nn.Conv3d(25, 64, kernel_size=3, padding=1),\n torch.nn.ReLU(inplace=True),\n )\n\n def forward(self, x):\n x = self.convs(x)\n x = self.interpolation(x)\n return x\n ```\n- Choose simply the **grid alignment** you like.\n \n ```python\n points, power, channels = 6, 3, 25\n interp = interp(points, power, channels, dtype=torch.double,align_corners = False)\n ```\n <p align=\"center\">\n <img src=\"https://github.com/ThomasHelfer/HigherOrderInterpolation3DTorch/blob/main/img/no_align.png\" alt=\"no alignment\" width=\"50%\" height=\"auto\"/>\n </p>\n or if you do not want to have any aligment with the input grid\n \n ```python\n points, power, channels = 6, 3, 25\n interp = interp(points, power, channels, dtype=torch.double,align_corners = True)\n ```\n \n <p align=\"center\">\n <img src=\"https://github.com/ThomasHelfer/HigherOrderInterpolation3DTorch/blob/main/img/align.png\" alt=\"aligned\" width=\"50%\" height=\"auto\"/>\n </p>\n\n- Choose the enhacement **factor** you like\n ```python\n factor = 4\n points, power, channels = 6, 3, 25\n interp = interp(points, power, channels, dtype=torch.double,align_corners = False,factor = factor)\n ```\n <p align=\"center\">\n <img src=\"https://github.com/ThomasHelfer/HigherOrderInterpolation3DTorch/blob/main/img/interpolation_grid_zoomed_factor_4.png\" alt=\"no alignment\" width=\"50%\" height=\"auto\"/>\n </p>\n \n ```python\n factor = 16\n points, power, channels = 6, 3, 25\n interp = interp(points, power, channels, dtype=torch.double,align_corners = False,factor = factor)\n ```\n <p align=\"center\">\n <img src=\"https://github.com/ThomasHelfer/HigherOrderInterpolation3DTorch/blob/main/img/interpolation_grid_zoomed_factor_16.png\" alt=\"no alignment\" width=\"50%\" height=\"auto\"/>\n </p> \n \n### Prerequisites\n\nBefore installing PyInterpX, ensure you meet the following prerequisites:\n- Python 3.8 or higher\n- pip package manager\n\n## License\n\nPyInterpX is open-sourced under the MIT License. For more details, see the LICENSE file.\n\n## Contact\n\nFor inquiries or support, reach out to Thomas Helfer at thomashelfer@live.de.\n",
"bugtrack_url": null,
"license": "MIT",
"summary": "A highly performant, GPU compatible package for higher order interpolation in PyTorch",
"version": "0.2.3",
"project_urls": null,
"split_keywords": [
"machine learning",
"pytorch",
"interpolation",
"higher order interpolation",
"3d interpolation",
"gpu compatible"
],
"urls": [
{
"comment_text": "",
"digests": {
"blake2b_256": "9d762ed3887e9a90afa57d6400d92b2900b2817eeea96649d17070b47f221af5",
"md5": "307ab3b9841f36068378ad28142b5185",
"sha256": "e5241e901825873122a58abfbcea3f9c90b830665b1f7a0eb333934df680b519"
},
"downloads": -1,
"filename": "PyInterpX-0.2.3-py3-none-any.whl",
"has_sig": false,
"md5_digest": "307ab3b9841f36068378ad28142b5185",
"packagetype": "bdist_wheel",
"python_version": "py3",
"requires_python": ">=3.8",
"size": 8329,
"upload_time": "2024-03-12T20:52:40",
"upload_time_iso_8601": "2024-03-12T20:52:40.734195Z",
"url": "https://files.pythonhosted.org/packages/9d/76/2ed3887e9a90afa57d6400d92b2900b2817eeea96649d17070b47f221af5/PyInterpX-0.2.3-py3-none-any.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "dd12e0f3d56fab2c80186fea50b15e7a4b6e9b807a04aedbb5791a8160bfd2c8",
"md5": "39598b4a9c65f347a17b3a8871fccf9a",
"sha256": "2667eb94b8cea6e6ffbd57797a31212ea7a096fbd45283732caf058253428438"
},
"downloads": -1,
"filename": "PyInterpX-0.2.3.tar.gz",
"has_sig": false,
"md5_digest": "39598b4a9c65f347a17b3a8871fccf9a",
"packagetype": "sdist",
"python_version": "source",
"requires_python": ">=3.8",
"size": 9737,
"upload_time": "2024-03-12T20:52:42",
"upload_time_iso_8601": "2024-03-12T20:52:42.603029Z",
"url": "https://files.pythonhosted.org/packages/dd/12/e0f3d56fab2c80186fea50b15e7a4b6e9b807a04aedbb5791a8160bfd2c8/PyInterpX-0.2.3.tar.gz",
"yanked": false,
"yanked_reason": null
}
],
"upload_time": "2024-03-12 20:52:42",
"github": false,
"gitlab": false,
"bitbucket": false,
"codeberg": false,
"lcname": "pyinterpx"
}
JSON