Software
parallel-video-io
A focused Python library for reading and writing videos with sane defaults, and for loading collections of videos efficiently in parallel — returning batches of frames in a single iterator, useful for machine learning inference in production.
- Built on TorchCodec, but 2–5× faster in the targeted use cases
- Smart load balancing across workers
- Unified API for real videos or directories of individual image frames
parallel-matplotlib-animation
Create matplotlib animations rendered to video in parallel, with efficient resource reuse. Matplotlib’s built-in animation module is single-threaded, and existing parallel solutions use embarrassingly parallel workloads without resource sharing. This library uses multiple processes with resource reuse for ~1.5–2× speedup, scaling near-linearly up to the number of CPU cores.
NeuroMechFly: fruit fly simulation in MuJoCo
Documentation · GitHub (FlyGym) · Workshop recording · Outreach game
A framework for simulating embodied sensorimotor control in adult Drosophila melanogaster, built in MuJoCo. NeuroMechFly provides a physically accurate fly model with realistic body, legs, wings, and sensory systems — enabling research at the intersection of neuroscience, biomechanics, and robotics.
A workshop on NeuroMechFly was held at the Society for Neuroscience (SfN) 2024 meeting. The associated paper was published in Nature Methods.