Research program
Visual and audio information.
Describing images and sound in exact, repeatable structures that other systems can rely on.
Proven in internal tests
What works today.
-
16,777,216
colours matched exactly
Fast paths match the reference exactly.
The accelerated image paths gave byte-identical results to the reference path for every possible 24-bit colour.
Internal certification run, one machine. Synthetic and test images only.
-
573 / 573
image checks passed
Image-structure components pass in full.
25 of 25 test suites passed in release and debug builds, and three image styles reproduce their reference outputs byte for byte.
Internal certification run. Not yet evaluated on real photographs.
-
213 / 213
audio architecture tests
Audio foundation in place.
The audio research architecture passed every test in release and debug builds, with zero warnings.
Internal certification run. Architecture only: no speech, music or audio analysis yet.
Images
Exact, repeatable image structure.
Describe the structure of an image the same way every time, so later systems can rely on it.
Next: results on real-world photographs.
Sound
An audio foundation, ready to build on.
The architecture for representing sound and speech is in place and fully tested.
Next: the first audio measurements. No speech recognition, speech comprehension or audio analysis capability is claimed.
Milestones
Every gate, and where it stands.
-
Deterministic image-structure components
Software components that describe the structure of an image in a repeatable, deterministic way.
Evidence and next gate
Shown: Core components pass internal automated tests on synthetic test images. A deterministic image stylization component passes internal tests.
Not yet shown: Visual understanding of any kind. Results on real-world image collections.
Next gate: The remaining defined behaviours implemented and tested.
-
Audio and speech research architecture
The defined structure of a future audio and speech research system.
Evidence and next gate
Shown: An architectural skeleton exists and passes internal structural tests.
Not yet shown: Speech recognition or comprehension of any kind. Audio analysis results.
Next gate: The remaining undefined parts of the architecture specified.