Work · Agricultural vision
Potato leaf transfer and robustness
- Demo ready
- Synthetic demonstration
- Related literature only
Transfer learning with a controlled corruption suite, temperature calibration and a model card exported from the actual run manifest.
Problem
Synthetic corruptions are not field validation, but they show where a classifier's confidence falls apart.
Contribution
Relationship to research: Related literature only. Cites a potato-leaf classification paper as related literature only; independent implementation.
Account owner’s role: Commissioned, reviewed and published the implementation. The code was produced with AI-assisted generation on commission and reviewed before release; it is not the original code of any cited publication.
What was built
- Local-folder adapter with provenance manifest
- Noise, blur, brightness, quantisation and occlusion suites
- Temperature scaling on validation
- Model card export
Method
Evaluate refuses to run without a provenance manifest for the data.
Data and access
No dataset bundled. PlantVillage is a candidate source whose licence grant was not established; the user must verify the release's permission.
Evaluation protocol
Per-class accuracy and macro F1 per corruption and severity, from real model outputs.
Results
No measured result is published for this project. Demonstration outputs from the repository’s own synthetic fixtures are labelled as such in the repository and are not presented here as results.
Limitations
- No benchmark claim exists until a licensed dataset and its class map are in place.
Reproducibility
Offline synthetic smoke; corruption functions unit-tested.
Repository metadata: Python, MIT, last push 2026-09-13, 0 stars.