Okuafo MaizeGuard Edge v1.4

00 / Overview

Two-stage offline maize screening system

A research-preview, two-stage LiteRT and TensorFlow Lite vision system evaluated on documented development datasets and limited field samples for offline, human-reviewed maize-leaf screening.

01 / Status

Public access and evidence basis.

Public repository visibility identifies access, not production approval or an open-source license grant. Review the current repository terms before copying, deploying, redistributing or modifying the artifacts.

02 / Intended capabilities

Published workflow and intended capabilities.

• Screen maize-leaf images as diseased, healthy or unsuitable for screening
• Route suspected disease to a focused five-category pattern screen
• Run without continuous network access on compatible mobile or edge hardware
• Return structured class probabilities and clear uncertainty pathways
• Support human-reviewed field scouting and agronomy workflows

03 / Research objective

Research objective

Make focused agricultural vision useful in low-connectivity settings while preserving rejection, uncertainty and human review as part of the screening workflow.

04 / Model files

Model files

• Stage 1: 24.5 MiB FP16 TensorFlow Lite triage gate
• Stage 2: 24.5 MiB FP16 TensorFlow Lite disease-pattern screen
• Machine-readable metadata and SHA-256 checksums
• Reference Python inference entry point

05 / Inputs

Inputs

• 384 x 384 RGB maize-leaf image
• Float32 tensor with raw RGB values in the 0-255 range

06 / Outputs

Outputs

• Stage 1: diseased, healthy or not maize / unclear
• Stage 2: bacterial leaf streak, common rust, gray leaf spot, northern leaf blight or other disease
• Class probabilities, routing decision and human-readable screening result

07 / Architecture overview

Architecture overview

Both stages use an EfficientNetV2B3 backbone with FP16-compressed weights. The first stage acts as a relevance and health gate; the second runs only when the gate supports focused disease screening.

08 / Limited field-sample basis

Limited field-sample basis

Limited field samples exercised the packaged two-stage, on-device workflow and informed image preprocessing, rejection behavior, confidence routing and the focused screening flow. Those samples were not an independent, agronomist-labelled field evaluation and do not support field-accuracy or diagnostic claims.

09 / Data and provenance

Data and provenance

Training images are not distributed with the model. The Hugging Face repository contains the current source citations, acknowledgements and provenance notes; access to the model does not grant rights to third-party source data.

10 / Known limitations

Known limitations

• Visible disease patterns may overlap with nutrient stress, chemical injury, weather damage, pests, senescence or mixed infections
• Blur, glare, shadows, occlusion, background clutter and early symptoms can cause rejection or incorrect screening
• Performance can vary by camera, farm, region, season, cultivar and crop stage
• The other-disease output is an uncertainty pathway, not a confirmed diagnosis

11 / Safety considerations

Safety considerations

Present outputs as screening results with confidence and uncertainty preserved. Treatment, pesticide, insurance, credit or other consequential decisions require review by a qualified agronomist, extension officer or appropriate specialist.

12 / Intended use

Intended use

Offline, human-reviewed maize-leaf screening; field-scouting triage; agronomy workflow support; and research on compact agricultural vision systems.

13 / Out-of-scope use

Out-of-scope use

Autonomous diagnosis, treatment selection, general-purpose crop classification, or consequential agricultural and financial decisions without specialist review.

14 / Access and licensing

Access and licensing

The repository and model files are publicly accessible on Hugging Face. Copying, deployment, redistribution and derivative use remain governed by the license file published in that repository; public access is not an open-source license grant.

15 / Citation

Citation

Sinuosity AI. Okuafo MaizeGuard Edge v1.4. 2026. Research-preview two-stage TensorFlow Lite maize screening system.

16 / Version history

Version history

Version 1.4 was published on Hugging Face on 21 July 2026 with two FP16 model artifacts, metadata, inference code, checksums, citations and responsible-use guidance.

17 / Responsible disclosure

Responsible disclosure

Potential safety, provenance, licensing or security concerns may be reported to hello@sinuosity.ai.