← All modules
MOD-08
Neural Vision Edge
Train, quantise and deploy vision models that run on edge silicon — detection, tracking and anomaly alerts with sub-frame latency and no cloud round-trip.

INFERENCE
7ms
Starting at
$45
Tiers
3
Free Tier Sandbox
Free tier sandbox — Edge inference viewer
Visitor · 0 XP
A simulated camera feed running detection and tracking in the browser. Free tier previews the pipeline — connecting a real stream, fine-tuning or deploying to edge silicon needs a paid tier.
next: Tinkerer
Model
EDGE-S
Latency
7 ms
Objects
0
Tracking
ON
Backbone
EDGE-S
Throughput
143 FPS
mAP@0.5
0.86
Threshold
0.50
| Class | Precision | Recall | F1 |
|---|---|---|---|
| person | 0.895 | 0.870 | 0.882 |
| forklift | 0.878 | 0.858 | 0.868 |
| pallet | 0.861 | 0.846 | 0.853 |
| vehicle | 0.844 | 0.834 | 0.839 |
| helmet | 0.827 | 0.822 | 0.824 |
| spill | 0.810 | 0.810 | 0.810 |
Generated pipeline config
preprocess:
resize: [640, 640]
color_space: RGB
normalize: { mean: [0.485, 0.456, 0.406], std: [0.229, 0.224, 0.225] }
letterbox: true
backbone:
name: EDGE-S
stages: [stem, C2f x3, SPPF]
precision: int8
head:
type: anchor-free decoupled
nms_iou: 0.45
conf_threshold: 0.50
tracker: bytetrackTier Selection
Tier II
Fovea
$189
per month
- Up to 12 streams
- Custom model fine-tuning
- Edge quantisation + deploy
- Webhook + alert pipeline
Tier III
Cortex
$549
per month
- Unlimited streams
- On-prem inference cluster
- Continuous retraining loop
- Dedicated vision engineer