Modelpublic

Bittensorminingfactory/streetvision-roadwork-v2

sourceHugging Faceupdated 7mo agoView on Hugging Face

0likes5downloads

Model Card

StreetVision Roadwork Detection Model (Binary)

Binary-compatible FasterViT model for SN72 StreetVision subnet.

Model Details

—Architecture: FasterViT-0 with binary output wrapper
—Output: Single float [0.0, 1.0] indicating roadwork presence
—Input: 224x224 RGB images
—Classes: D00, D10, D20, D40 (internally mapped to binary)

Usage

python

from binary_wrapper import load_model

model = load_model("pytorch_model.bin", method="max")
score = predict(model, "image.jpg")  # Returns float [0,1]

Binary Conversion

The 4-class damage detection model outputs probabilities for:

—D00: Longitudinal Crack
—D10: Transverse Crack
—D20: Alligator Crack
—D40: Pothole

Binary score = max(probabilities) across all damage classes.