NERO: Non-Equivariance Revealed on Orbits

Code Repository of our NERO Plots paper

Getting Started

A virtual environment is recommended for installing dependencies, e.g., using conda:

conda create --name nero_env python=3.9

and then

conda activate nero_env

Next we can install all the dependencies, which are summarized in qt_app/setup_env.sh.

bash qt_app/setup_env.sh

NERO Interface

After installation, you should be able to run the NERO Interface by

python qt_app/nero_app.py --mode digit_recognition --demo

nero_app.py takes three arguments:

• --mode: Initialize the interface for different use cases. Currently it supports digit_recognition, object_detection and piv. But more to come. Please feel free to create a pull request for new interface.

• --cache_path: NERO Interface does computations in realtime and saves the results to cache_path that could be loaded next time during initialization. It saves time when users want to re-examine NERO plots that they created before. Can leave as open by default, but can also define a specific path that leads to a specific cache. One example use case could be that you have different versions of the same model that work all in one mode (digit_recognition, object_detection and piv).

• --demo: A binary flag that defines the behavior of NERO Interface. Without the flag, NERO Interface will be running in developing fashion that helps debugging. Users should always include this flag.

Digit Recognition

As demoed in the Method section in our paper, digit recognition task can be visualized within NERO Interface by running

python qt_app/nero_app.py --mode digit_recognition --demo