MATLAB® based Python® Requirements manager

MPyReq is a MATLAB-based Python requirements manager that simplifies installing and configuring Python environments for MATLAB projects.

MPyReq is a MATLAB-based Python requirements manager that simplifies installing and configuring Python environments for MATLAB projects. It handles Python version management, package installations, git repository cloning, and weight file downloads with a simple API. MPyReq ensures consistent Python environments across sessions by caching configurations and providing runtime setup for your Python dependencies.

Setup

Call MPyReq.setInstallFolder(largeSSDFolder) to set the install folder.

This only needs to be done once per machine. All subsequent uses of MPyReq share downloaded content from this folder.

Getting Started

Invoke help to see documentation

help MPyReq.m

Note: Click 'Open in MATLAB Online' and run the demoCellpose example in MATLAB Online.

You can set install folder to "/tmp/pydata" when trying it out.

Example

% First ever run of this section     - Does installation and Runtime setup
% First run in a subsequent session  - Just Runtime Setup
% Second run in any session          - A very quick check to ensure env is up
MPyReq.python("3.12");
MPyReq.pipPackage("cellpose"); 


im = imread("AT3_1m4_01.tif");

% Use Python interface to call into Python code
% See: https://cellpose.readthedocs.io/en/latest/api.html
model = py.cellpose.models.CellposeModel(gpu=true);
outputs = model.eval(im);

labels = uint16(outputs{1});
imageshow(im,OverlayData=labels)

More Examples

Run the following examples after ensuring to set an installation folder using the call below. Ideally, this folder should be a fast access (e.g SSD) with ~15+ GB free space. These examples download large libraries and model files.

These examples were tested on a glnxa64 machine with a working GPU.

Ensure to run this to clear the python environment between two demo runs:

terminate(pyenv); clear MPyReq

demoCellpose - Run https://www.cellpose.org/ cell segmentation model.

demoStarDist - Run https://github.com/stardist/stardist object detection with star convex shapes

demoDepthPro - Run https://github.com/apple/ml-depth-pro A quick, sharp monocular metric depth model.

demoSAM2 - Run https://github.com/facebookresearch/sam2 Segment Anything Model 2

demoVGGT - Run https://vgg-t.github.io/ Visual Geometry Grounded Transformer

Troubleshooting

Python version or Package Conflicts

MPyReq does not resolve package conflicts. If you tried to install Package A first which used x.xx version of an upstream package U, and then tried to add Package B which used y.yy version of U in the same pyenv session - you will get odd Python errors due to conflicts in required version of Package U.

Best strategy is to work with one project at a time, and clear the environment before switching projects with conflicting requirements.

terminate(pyenv); clear MPyReq

Workflows needing models with conflicting requirements will have to serialize their runs with the above reset between calls. For best performance, run all data through model 1 first, serialize the results into disk (or MATLAB variables if possible), and then run model 2 on the saved/cached data.

win64

MPyReq is expected to work on win64.

However, installing CUDA enabled PyTorch has not been tested. MPyReq uses uv as the install engine, so uv environment variables are honored. You can experiment with setting this environment variable before invoking any installs to guide uv to download from the pytorch index

setenv("UV_EXTRA_INDEX_URL","https://download.pytorch.org/whl/cu126")

Alternate repos

If you have a company internal repo, use uv environment variables, like below to include that in MPyReq workflows.

setenv("UV_DEFAULT_INDEX", "your internal repo")