Technical brief
TRIKS · Trimodal Kinematic‑Spatial learning
A radar model
that learns without labels.
Partners pretrain on their own unlabelled radar corpora and fine-tune only the last mile.
01 · Recipe
Self-supervised multi-representation learning.
Proprietary. 4 US patents pending.
02 · Variants
| Variant | Assumption | Status |
|---|---|---|
| Cross-modal (RV) | Paired radar & camera on the same platform | Validated |
| Radar-only (R) | Radar alone; temporal self-supervision; camera‑less platforms | Piloting |
The radar-only variant enables maritime, counter-UAS and Arctic surveillance where cameras may not be available or appropriate.
03 · Benchmark
Performance characterisation on public datasets for reproducibility and comparison.
Available under NDA.
04 · Provenance
Earlier incarnations were published at ICC 2022, CVPR 2023, and CVPR 2024. Four US patents pending.
Start a pilot info@radareye.ai