Neural Sensing Engine™

Replay your unique sensor stack against our corpus.

We remap the capture onto the payload you ship and run the stack you will actually field. Completely offline.

Licensed models

UAV Detection Model

Licensed detection remapped to the payload you ship.

mAP@0.5
94.2%
Inference
120 fps
Model size
47 MB

UAV Tracking Model

Single-object tracking on the same remap. Holds the airframe across the trajectory.

MOTA
87.4%
Success
91.2%
P95 latency
18 ms

RL Environments

The corpus is the environment. You bring the policy.

Observation

Frame
01482
N (m)
+012.40
E (m)
-004.18
D (m)
+002.10

Scorecards

Where the detection and tracking stack will hold, and where it will fail. Group 1, Group 2, Group 3. Completely offline.

Offline Benchmarking

Measured

Result

Measured

Offline Benchmarking completed. This report makes no System Qualification decision.

Hash-verified artifact bundle

7c3a91e0b2f14d8c

UAV Detection Model · Hill Country Day

Type

model benchmark

Completed

Aug 11, 2026, 2:14 PM

Release

hill-country-day-v1

Model

vigil-association-transformer-v1

Split

held-out

Spectrum

RGB

Detection Evaluation

12,480 frames

mAP@[.50:.95]0.742
AP500.942
Calibration ECE0.0410
P95 latency18.00 ms

Single-Object Tracking (SOT) Evaluation

8,112 scored frames

Success0.874
Precision0.912
Normalized precision0.889
State accuracy0.931