Mirror Fractal Codec

Lossless compression for neuromorphic event-camera data (DVS/ATIS).

1.51 bits/event · 47.6x compression · 0.55 ms per N-MNIST sample.

Patent Pending — US Provisional #64/034,974

Install

pip install mfc-codec

Pre-built wheels for CPython 3.11 on Linux (manylinux_2_34 x86_64), macOS (arm64), and Windows (amd64).

Quick start

import mfc

events = mfc.load_nmnist("sample.bin")
compressed, stats = mfc.compress(events, width=34, height=34, t_bits=16, xy_bits=8)
print(f"{stats.compression_ratio:.1f}x, {stats.bits_per_event:.2f} bpe")

decoded = mfc.decompress(compressed)
assert events == decoded   # bit-exact, lossless

Results

N-MNIST (60,000 real samples, 34×34 sensor)

High-resolution synthetic

Supported dataset formats

N-MNIST .bin · AEDAT 2.0 .aedat · DSEC HDF5 (Blosc/LZ4/zstd) · .npy · .evb · CSV · .dat/.raw

Links

• Live demo — https://codec.mirrorfractal.com
• Public repository (docs, examples, issues) — https://github.com/MirrorFractal/codec
• Paper / Zenodo deposit — DOI: 10.5281/zenodo.19704064

Citation

@software{mfc_codec_2026,
  author = {Solonskii, Aleksei},
  title  = {Mirror Fractal Codec: Lossless Compression for Neuromorphic Event Camera Streams},
  year   = {2026},
  doi    = {10.5281/zenodo.19704064},
  url    = {https://codec.mirrorfractal.com}
}

License

Proprietary — all rights reserved. Patent Pending US 64/034,974.

Contact: info@mirrorfractal.com