Systems and low-level
A from-scratch Rust operating system with capability security and deterministic replay, and an intent runtime whose formal proofs run in microseconds.
GENOME
A content codec built from nothing, targeting 100:1.
Research Compression research
GENOME turns a file into a tiny genome and regenerates it using only compute — no stored slice of the original, no re-fetching, and no off-the-shelf compressor anywhere in the pipeline.
What it does
Everything in it is hand-built: the rANS entropy coder, the 8×8 DCT transform, quantisation, the coefficient model and the motion search. No compression library is used at any stage. Standard image containers are read and written for input and output only, and ffmpeg is used as a raw-pixel pipe rather than as a codec.
It exists in two implementations, Python and Rust, which is what makes the results checkable: the same input has to produce the same genome in both, and a bug in one shows up as a disagreement rather than as a plausible output.
Research, and honest about it. The target is a hundred to one and the results file records where it actually lands rather than where we would like it to.
Who it is for
Compression and codec researchers, and anyone who wants to see a codec built from first principles rather than configured.
What it costs
Research project.
Terminology
Words this page uses in a specific sense.
- rANS
- Range asymmetric numeral systems — an entropy coder with arithmetic-coding compression ratios at close to Huffman speed.
- DCT
- Discrete cosine transform. Converts a block of pixels into frequency coefficients, most of which are small enough to discard cheaply. It is the transform underneath JPEG.
- Motion search
- Finding where a block of one video frame moved to in the next, so the codec stores the movement rather than the pixels again.
Facts
| Field | Value |
|---|---|
| Name | GENOME |
| Status | Research |
| Category | Compression research |
| Platforms | Windows, Linux, macOS |
| Pricing | Research project. |
| Source | Closed source |
| Publisher | Cognitive Industries |
| Published from | Brisbane, Queensland, Australia |
Capabilities behind it
Computer vision
Real-time GPU pipelines: licence plates through extreme degradation, FaceID with persistent tracking, fire and smoke detection.
Media and streaming
Multi-tenant live streaming with automated capture, encoding and scoreboard overlays; in-browser file tooling where nothing is uploaded.
Related
Last reviewed · Site changelog