PIVPy PIVPy

Python-based post-processing PIV data analysis in the repo: https://github.com/openpiv/pivpy

PyPI version Documentation Status Binder

Merging the three packages:

  1. https://github.com/tomerast/Vecpy

  2. https://github.com/alexlib/pivpy/tree/xarray

  3. https://github.com/ronshnapp/vecpy

How do I get set up?

Recommended: use uv (fast, reproducible)

Create a virtualenv and install:

uv venv
uv pip install pivpy

Install this repository (editable / development install):

uv venv
uv pip install -e .

Install with optional dependencies (including lvpyio for LaVision VC7):

uv pip install 'pivpy[full]'

Run in a sandbox (isolated) environment (no persistent venv required):

uv run --isolated --with pivpy python -c "import pivpy; print(pivpy.__version__)"

Or run with optional dependencies in the sandbox:

uv run --isolated --with "pivpy[full]" python -c "from pivpy import io; print(io)"

Run using a specific Python version (uv-managed) in the sandbox:

uv python install 3.14
uv run --isolated --managed-python -p python3.14 --with pivpy python -c "import sys, pivpy; print(sys.version.split()[0], pivpy.__version__)"

Run this repository’s tests on Python 3.14 (sandboxed):

uv run --isolated --managed-python -p python3.14 --with-editable . pytest -q

Alternative: use pip:

pip install pivpy

or with optional dependencies:

pip install 'pivpy[full]'

if you use OpenPIV, PIVlab, etc.

Quick start (auto-detect file format):

from pivpy import io
ds = io.read_piv('your_file.vec')

Getting Started

Load a built-in sample dataset, compute vorticity, and plot velocity vectors (quiver) over a vorticity heatmap:

import numpy as np
import matplotlib.pyplot as plt
from scipy.ndimage import gaussian_filter

import pivpy.pivpy  # registers the .piv accessor
from pivpy import io

# Built-in sample data shipped with pivpy (OpenPIV vector file)
ds = io.read_piv('pivpy/data/openpiv_vec/exp1_001_b.vec').isel(t=0)

# Vorticity (stored as ds['w'])
ds = ds.piv.vorticity(name='w')

X, Y = np.meshgrid(ds['x'].values, ds['y'].values)

fig, ax = plt.subplots(figsize=(8, 4.8))

# Background: smoothed vorticity heatmap (robust symmetric color limits)
w = ds['w'].values
w_smooth = gaussian_filter(w, sigma=1.0)
finite_w = w_smooth[np.isfinite(w_smooth)]
vmax = np.nanpercentile(np.abs(finite_w), 98) if finite_w.size else 1.0
vmax = max(vmax, 1e-9)
pcm = ax.pcolormesh(X, Y, w_smooth, shading='auto', cmap='coolwarm', vmin=-vmax, vmax=vmax)
fig.colorbar(pcm, ax=ax, pad=0.02, label='vorticity (smoothed, 1/Δt)')

# Overlay: velocity vectors (make arrows longer by reducing `scale`)
skip = 2
ax.quiver(
    X[::skip, ::skip],
    Y[::skip, ::skip],
    ds['u'].values[::skip, ::skip],
    ds['v'].values[::skip, ::skip],
    color='k',
    angles='xy',
    scale_units='xy',
    scale=5.0,
    width=0.004,
)

ax.set_title('PIVPy sample (OpenPIV vec): quiver over smoothed vorticity')
ax.set_xlabel('x')
ax.set_ylabel('y')
ax.set_aspect('equal')
fig.tight_layout()
plt.show()

Quiver over vorticity heatmap

Legacy loaders (still supported):

ds = io.load_vec('your_file.vec')
ds = io.load_openpiv_txt('your_file.txt')

Check whether a newer version is available on PyPI:

import pivpy
res = pivpy.check_update(verbose=True)
# res.status: 0=unavailable, 1=up-to-date, 2=update available, 3=installed newer

PIVMat-inspired methods

PIVPy exposes many post-processing operations via the xarray accessor Dataset.piv. Several common PIVMat toolbox methods are available with similar names/behavior:

import pivpy.pivpy  # registers the .piv accessor
from pivpy import io

ds = io.create_sample_Dataset(n_frames=10)

# Add noise (similar to PIVMat addnoisef)
ds_noisy = ds.copy().piv.addnoisef(eps=0.1, opt='add', nc=0.0, seed=0)

# Ensemble (temporal) average and optional std/rms (similar to PIVMat averf)
avg = ds.piv.averf()
avg, std, rms = ds.piv.averf(return_std_rms=True)

# Spatial averages (similar to PIVMat spaverf)
ds_xy = ds.piv.spaverf('xy')   # excludes zeros by default
ds_x0 = ds.piv.spaverf('x0')   # include zeros

# Subtract ensemble/spatial average (similar to PIVMat subaverf)
fluct = ds.piv.subaverf('e')
fluct_x = ds.piv.subaverf('x0')

# Azimuthal averaging (similar to PIVMat azaverf)
r, ur, ut = ds.isel(t=0).piv.azaverf(0.0, 0.0, return_profiles=True)

# Temporal resampling and phase average (similar to PIVMat resamplef/phaseaverf)
ds_r = ds.piv.resamplef(tini=range(ds.sizes['t']), tfin=[0.5, 1.5, 2.5])
phased = ds.piv.phaseaverf(12)

Additional PIVMat-inspired utilities:

# Correlation along a dimension (similar to PIVMat corrm/corrx)
cu = ds.piv.corrm(variable='u', dim='x')        # returns DataArray with a 'lag' dimension
cv = ds.piv.corrm(variable='v', dim='y', half=True)

# Spatial correlation function + integral scales (similar to PIVMat corrf)
cor = ds.piv.corrf(variable='u', dim='x', normalize=True)
# correlation curve: cor['f'] over cor['r']
# integral scales: cor['isinf'], cor['is5'], cor['is2'], cor['is1'], cor['is0']

# Fill holes encoded as zeros (similar to PIVMat interpolat.m behavior)
ds_filled = ds.piv.fill_zeros(max_iter=10)

# Extract a rectangular region (similar to PIVMat extractf)
sub = ds.piv.extractf([0.0, 0.0, 10.0, 5.0], 'phys')   # [x1,y1,x2,y2] in physical units
sub = ds.piv.extractf([10, 5, 50, 40], 'mesh')         # 1-based mesh indices (MATLAB-like)

# Spatial convolution filter (similar to PIVMat filterf)
ds_smooth = ds.piv.filterf(1.0, 'gauss', 'same')   # keep same size
ds_smooth_valid = ds.piv.filterf(1.0, 'gauss')     # smaller (conv2(...,'valid') behavior)

# Flip field (similar to PIVMat flipf)
ds_lr = ds.piv.flipf('x')     # left-right mirror (negates u)
ds_tb = ds.piv.flipf('y')     # top-bottom mirror (negates v)
ds_xy = ds.piv.flipf('xy')    # both

# 2D Butterworth filter (similar to PIVMat bwfilterf)
ds_low = ds.piv.bwfilterf(filtsize=3.0, order=8.0, mode='low', trunc=True)
ds_high = ds.piv.bwfilterf(filtsize=3.0, order=8.0, mode='high')

# PIVMat-style option wrapper: opts can include 'high'/'low'/'trunc'
ds_high2 = ds.piv.bwfilterf_pm(3.0, 8.0, 'high', 'trunc')

# Batch processing over filename series (similar to PIVMat batchf)
# fun can be a callable (fun(ds, ...)) or an accessor method name (e.g. 'averf', 'bwfilterf')
from pivpy.io import batchf
results = batchf('pivpy/data/day2/day2a00500[0:5].T000.D000.P003.H001.L.vec', 'averf')

For developers, local use:

Using uv (recommended):

git clone https://github.com/alexlib/pivpy .
cd pivpy
uv venv
uv pip install -e .

Editable install with optional dependencies:

uv pip install -e '.[full]'

Alternative (conda):

git clone https://github.com/alexlib/pivpy .
cd pivpy
conda create -n pivpy python=3.11
conda activate pivpy
conda install pip
pip install -e .

What packages are required and which are optional

  1. lvpyio by Lavision Inc. if you use vc7 files

  2. netcdf4 if you want to store NetCDF4 files by xarray

  3. pyarrow if you want to store parquet files

  4. vortexfitting if you want to do vortex analysis ($\lambda_2$ and $Q$ criterions, vortex fitting)

  5. numpy, scipy, matplotlib, xarray are must and installed with the pivpy

Contributors

  1. @alexlib

  2. @ronshnapp - original steps

  3. @liorshig - LVreader and great visualizaiton for Lavision

  4. @nepomnyi - connection to VortexFitting and new algorithms

How to get started?

Look into the getting started Jupyter notebook

and additional notebooks: Notebooks

How to test?

From a command line just use:

pytest

With uv:

uv run pytest -q

Documentation on Github

PIVPy on ReadTheDocs

How to help?

Read the ToDo file and pick one item to program. Use Fork-Develop-Pull Request model to contribute

How to write tutorials and add those to the documentation

Using great tutorial http://sphinx-ipynb.readthedocs.org/en/latest/howto.html we now can prepare IPython notebooks (see in /docs/source) and convert those to .rst files, then

uv pip install -r docs/requirements.txt
uv run sphinx-build -b html docs/source/ docs/build/html

generates docs/build/html directory with the documentation

PIVPy

Indices and tables