jnormcorre

This is a Jax-accelerated implementation of normcorre.

Installation

Currently, this code is supported for Linux operating systems with python version at least 3.8. We primarily use JAX for fast CPU/GPU/TPU execution and Just-In-Time compilation; see the Google installation instructions on how to install the appropriate version of JAX and JAXLIB for your specific hardware system. We require:

python>=3.8
jax>=0.3.25
jaxlib>=0.3.25

To install this repository into your python environment from the source code, do the following (this approach is recommended for now):

#Step 1: Install the appropriate version of jax for your hardware system 

#Step 2: Run the below lines of code
git clone https://github.com/apasarkar/jnormcorre.git
cd jnormcorre
pip install -e .

To install the most recently published version from PyPI, you can do:

#Step 1: Install the appropriate version of jax for your hardware system 

#Step 2: Run below line
pip install jnormcorre

The package on PyPI comes with jax, but doing the install in this order will allow you to control whether your version of jax is GPU/TPU compatible for your system. If you are only running on CPU, you can just skip to step 2.

Use Cases

This implementation can support both online and offline motion correction use cases. Here are some common ones:

1. Offline: Given a full video, write out the motion corrected video as a new file.

2. Offline + PMD Compression&Denoising: Given a full video, estimate templates, and then perform on-the-fly registration (rigid and/or piecewise rigid) + compression. No need to save an intermediate registered movie - instead, reduce your data size by ~2 orders of magnitude and use this for downstream processing tasks. This is hyper efficient thanks to jax's composability mechanisms, and useful for large data analysis, online experiments and more. This setup is implemented in PMD and in the full maskNMF functional imaging analysis pipeline.

3. Online: jnormcorre can be set up to adaptively estimate a template, and then take a single pass through a stream of new data, registering all newly observed frames and updating the template in the process

Citations

• Eftychios A. Pnevmatikakis and Andrea Giovannucci, NoRMCorre: An online algorithm for piecewise rigid motion correction of calcium imaging data, Journal of Neuroscience Methods, vol. 291, pp 83-94, 2017; doi: https://doi.org/10.1016/j.jneumeth.2017.07.031

• Matlab Implementation of Normcorre: https://github.com/flatironinstitute/NoRMCorre#ref

• Python Implementation of Normcorre: https://github.com/flatironinstitute/CaImAn

License

See License.txt for the details of the GPL license used here.