Two separate AI models examined your leaf independently (an ensemble). When both pick the same disease, the diagnosis is more reliable.

If the badge in the middle says "disagree", treat the result as tentative and consider retaking the photo or asking an agricultural extension officer. Strong agreement is a strong trust signal.

Each dial is one AI model's answer to "how sure are you about the final diagnosis?" A fuller ring means that model was more confident.

The best case is both dials nearly full and coloured green. A big gap between the two dials is a quiet form of disagreement about how confident the ensemble should be.

Each bar is one possible disease. The top half of the bar shows how likely the v3 model thinks your leaf has that disease; the bottom half shows the same for the sp_lora model.

When both halves reach the same length, the two models agree on that class. When one half is visibly longer than the other, the models disagree, and the black tick in the middle is the averaged opinion (ensemble) that drives the final diagnosis.

Temperature is a setting that controls how "opinionated" the AI is. A temperature below 1 concentrates probability onto the top guess; a temperature of exactly 1 leaves the softmax scores unchanged.

The left column of bars shows v3's raw scores before the temperature adjustment. The right column shows them after. This calibration step is what makes the final diagnosis decisive instead of hedging across several classes.

Each row is one quality check that runs on your result. Green rows passed, amber rows were skipped. We start at the top and walk downward.

The boxed row at the bottom is the final tier. Tier 1A is the best (high confidence). Tier 4A means the system chose to abstain; if you see 4A, treat the diagnosis as a guess and retake the photo with better lighting or a closer angle.

The coloured bar shows three quality zones. Red on the left means "too uncertain to diagnose". Amber in the middle means "possible but not confident". Green on the right means "confident diagnosis".

The black pin is where your result actually sits. The deeper into the green zone the pin is, the higher the tier. If the pin is in the red zone, the system is abstaining; retake the photograph with better lighting or a closer angle.

These are all six tomato diseases the AI considered. The row highlighted in green is the one it picked as most likely. The percentage next to each row is how likely the AI thinks that disease is, out of 100%.

Reading the list from top to bottom gives you the ranking of possibilities. If the top row is far ahead of the second row, the AI is sure. If the top two rows are close, the diagnosis is a differential and deserves a second look.

Your leaf's appearance is turned into a compact numeric fingerprint. We compare it against about 30 fingerprints of training prototypes we already know the answers to. This chart plots all 31 points in a 2D map via PCA.

The pulsing dried-leaf silhouette is your leaf. The coloured leaves around it are prototypes grouped by disease. Leaves close to your silhouette are visually similar (high cosine similarity). Lines connect your leaf to the three closest matches. Below, the "Your leaf looks most like" strip shows those three training images side by side so you can eyeball the similarity yourself.