File size: 1,659 Bytes
8385016 1b0739e 8385016 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 | ---
license: agpl-3.0
library_name: onnxruntime
tags:
- bittensor
- object-detection
- ppe
- onnx
---
# SN44 SafetyVision PPE miner
This package targets `manak0/Detect-worksite-ppe-compliance-detection-probe-set`
on Score/TurboVision SN44. It wraps a 320px ONNX export of the SafetyVision v1
YOLOv8n model and maps its four relevant labels into the active manifest order:
1. safety helmet
2. high-visibility safety vest
3. protective gloves
4. protective eyewear
The runtime confidence threshold is 0.10, selected on the public
Construction-PPE proxy set using the subnet's released 60% mAP50 / 40%
false-positive formula. Proxy results are documented in
`../sn44-opportunity-2026-10-02.md`. They do not establish performance on the
hidden `synthetic_r2` challenges.
The 12.15 MB model was exported from the upstream v1 checkpoint at 320px. It
was tested with Ultralytics 8.3.190 and ONNX Runtime 1.26.0, the versions
resolved or pinned by the official TurboVision compliance container. Model source:
`ayushgupta7777/safetyvision-yolov8` (`best.pt`, v1). The model card states
that the weights inherit the Ultralytics YOLOv8 AGPL-3.0 license.
The miner uses direct NumPy/OpenCV preprocessing and postprocessing rather than
importing Ultralytics or PyTorch at runtime. It initializes ONNX Runtime in
sequential mode with one intra-op thread. This keeps it inside the compliance
worker's 16-process/thread resource limit as well as its two-vCPU latency
envelope.
This repository contains the immutable model package used by an SN44 mining
trial. Deployment and on-chain registration are operational state and are not
asserted by this model card.
|