Model Disclaimer

ChiliSense uses two machine-learning models in sequence:

1. YOLOv11-seg — an instance segmentation model that detects and outlines each chili pepper in an uploaded image.
2. EfficientNetV2-S — a binary classification model that labels each detected pepper as healthy or unhealthy.

Outputs are statistical predictions accompanied by confidence scores. They are not certified diagnoses.

The models were trained on a curated dataset of approximately 1,400 chili-plant images sourced from licensed public datasets and agricultural-research material. The dataset emphasises long, slender cayenne-style cultivars (e.g. cayenne, Thai bird's eye, Kashmiri). Coverage of other varieties (bell peppers, jalapeños, habaneros, ornamental cultivars) is comparatively limited, and the modest dataset size means predictions outside the trained distribution should be treated with extra caution.

ChiliSense is intended as a decision-support tool for:

• Small-scale and commercial growers who want a quick triage signal across fields.
• Agricultural-extension workers documenting crop status.
• Researchers exploring computer-vision tooling in agriculture.
• Hobbyists curious about their own chili plants.

The model's predictions can degrade or fail when:

• Lighting is poor (deep shadow, direct sunlight, night photography).
• Peppers are heavily occluded by leaves or other foliage.
• The image resolution is below ~1024 px on the long edge.
• The pepper variety differs substantially in shape or colour from cayenne-style training examples.
• Multiple peppers overlap in a single cluster (the count may be under-reported).
• The image contains motion blur, water droplets, or heavy compression artefacts.

The unhealthy classification is a coarse visual flag. It detects appearance changes consistent with disease, sunscald, pest damage, or severe ripening defects — it does not identify which condition is present.

ChiliSense does not provide agronomic, veterinary, medical, financial, or legal advice. Do not use the output as the sole basis for crop treatment, sale, disposal, or any decision with material consequences. Confirm findings with a qualified agronomist or extension specialist before acting.

We do not warrant that the model's output is accurate, complete, current, or fit for any particular purpose. Reported confidence scores reflect the model's internal certainty — they are not a guarantee of correctness in the real world.

By using ChiliSense you accept that you alone are responsible for any decisions, actions, or inactions you take based on the model's output. You also agree to indemnify ChiliSense and its operators against claims arising from your use of those outputs.

We periodically retrain and update the underlying models. Behaviour may change between releases — including detections becoming more or less aggressive. Major model changes are documented in product release notes.

• Terms of Service — overall rules for using the Service.
• Privacy Policy — how we handle your uploaded images and account data.
• API reference — the endpoints that expose the model's output programmatically.