A GPU-accelerated oblique interpolation library for oblique plane microscopy (OPM) volume reconstruction. Avoids aliasing and interpolation artifacts caused by rectilinear interpolation approaches.
Links: API documentation, GitHub repository
Oblique Plane Microscopy (OPM) acquires 3D volumes by scanning an inclined light sheet through the sample. Because the imaging plane is tilted relative to the camera axes, standard rectilinear interpolation during volume reconstruction can introduce aliasing artifacts and loss of resolution – problems that are ameliorated by oversampling, Fourier stitching or deconvolution (see McFadden et al. 2025 and Lamb et al. 2025).
opm_unshear performs true oblique interpolation in the sample’s native coordinate frame, mapping voxels along the tilted plane directly onto an isotropic grid (see Hoffmann et al. 2023, Equation 2). We achieve both artifact-free reconstructions and GPU-accelerated throughput.
- Oblique interpolation in oblique space, avoiding aliasing from misaligned sampling axes
- CUDA-accelerated compute kernel for real-time GPU processing (
opm_unshear.gpu.unshear, ~1 gigavoxel/s on a NVIDIA RTX 3090) - Numba-accelerated multi-threaded CPU fallback for non-GPU systems (
opm_unshear.cpu.unshear, ~0.1 gigavoxel/s on a 16-core CPU)
- Linux or Windows PC
- For GPU acceleration (recommended): CUDA-capable NVIDIA GPU with sufficient memory (≥ 10 bytes per voxel of the volumes to be processed) and drivers supporting CUDA version ≥ 11.2 (check with
nvidia-smi)
- Install dependencies using conda/mamba:
- clone or download environment.yml and type
conda create -n opm_unshear_env -f environment.yml(or if you already have a conda environment:conda env update -f environment.yml)
- Install opm_unshear:
conda activate opm_unshear_env(or the name of your conda environment)pip install opm_unshear(or, for development, clone this repository, change to the directory containingpyproject.tomland run:pip install -e .)
The oblique interpolation function requires the following parameters:
data: the input data, a 3D array of shape(p, v, h), wherepis the number of oblique planes (camera frames),vis the number of vertical camera pixels, andhis the number of horizontal camera pixels. (note that we are using numpy-standard C-order indexing, where the axes are ordered from slowest-varying to fastest-varying)slope(float): Imagining an ideal pencil parallel to the optical axis, this is the shift of the pencil image, invcamera pixels, between consecutive camera frames. It can be estimated in practice by imaging any structured sample and inspecting the apparent drift between camera frames. This parameter can be positive or negative, depending on the direction of plane scanning or oblique plane tilt. You can also useopm_unshear.get_slopeto calculate this value based on setup parameters.sub_j(int): subsampling factor along the vertical direction (axis 1, along the optical axis) of the output dataset. Values should be between 1 (no subsampling) andabs(slope).sup_i(float): supersampling factor along the plane-scanning direction (axis 0 of the output). Values should be between 1 andabs(slope)(higher values are possible but this just wastes memory).
Example geometry with slope=3, sub_j=2, sup_i=2 (showing a slice along axis 2, or "h"):
input output
o-----o-----o o--o--o--o--o
\ \ \ | | | | |
o-----o-----o | | | | |
\ \ \ | | | | |
axis 1 o-----o-----o --> axis 1 o--o--o--o--o
("v") \ \ \ | | | | |
o-----o-----o | | | | |
\ \ \ | | | | |
o-----o-----o o--o--o--o--o
axis 0 ("p") axis 0
import numpy as np
from opm_unshear import unshear, get_slope
slope, _, _ = get_slope(n1=1.33, n2=1, M12=1.6, M23=2, dv=5, dp=2, theta_sample=np.radians(30), polarity=1)
#polarity = 1 or -1, depending on the direction of plane scanning
data = np.random.rand(20, 30, 40).astype(np.float32)
result = unshear(data, sub_j=2, sup_i=2, slope=slope)python -m opm_unshear --input data.h5:dataset_name --output result.h5 --sub_j 2 --sup_i 2 --slope 5--input: path to the input file containing a 3D dataset to be deskewed (.h5, .mat, .nii, .npy). For hierarchical file formats (.h5 or .mat), specify the dataset name using the formatfilename.h5:dataset_name. Note that the dataset's axes have to be ordered as described above.--output: path to the output file (.h5)- all other parameters as described above
We value your feedback! If you encounter issues, have suggestions for improvements, or simply want to let us know that opm_unshear worked well for you, please reach out:
- Bug Reports: Open an issue on our GitHub Issues page.
- Feature Requests: If you have ideas for new features, feel free to suggest them via a GitHub issue.
We welcome contributions! Please:
- Follow the existing coding style and include tests for new features or fixes.
- Fork the repo, create a branch, and submit a pull request (PR) with a clear description.
If you use opm_unshear in your research, please cite the paper that first described our interpolation approach:
Hoffmann, M., Henninger, J. et al. Blazed oblique plane microscopy reveals scale-invariant inference of brain-wide population activity. Nature Communications 14, 8019 (2023). https://doi.org/10.1038/s41467-023-43741-x
@article{Hoffmann2023,
title={Blazed oblique plane microscopy reveals scale-invariant inference of brain-wide population activity},
author={Hoffmann, Maximilian and Henninger, Jorg and Veith, Johannes and Richter, Lars and Judkewitz, Benjamin},
journal={Nature Communications},
volume={14},
number={1},
pages={8019},
year={2023},
publisher={Nature Publishing Group},
doi={10.1038/s41467-023-43741-x}
}