Fruit Phenotyping Platform

To request access or submit feedback, please email zsong@uga.edu with your institution email.

Specialty Crop Fruit
Phenotyping Platform

Connecting to server...
Loading saved data...

Analysis Setup

Step 1: Select your fruit

This walkthrough configures your analysis step by step: select a fruit type, answer a few questions about your images, then review your setup. Once finished, move to Image Analysis to upload one or more images for processing. See the Help section at the bottom of this page for explanations of each feature.

Select Fruit is required before continuing.

Step 2: Scale and color calibration

? Print this sheet when you want a simple size reference in your images. Choose the marker size that fits your fruit, print it at the indicated physical size, cut it out if needed, and keep one marker flat and fully visible in each image. If a marker is detected, it can be used for automatic scale calibration.

Use an ArUco marker when you want automatic size calibration without entering a manual pixel-to-cm conversion. It is optional, but helpful when a known physical scale is needed.

Reference ColorChecker target: use the middle Passport page

ColorChecker reference: the ColorChecker is not necessary, but it is helpful for color calibration. Use the middle page of the Calibrite ColorChecker Passport Photo 2, the 24-patch page that looks like this reference image.

Scale priority: manual scale, then ArUco marker, then ColorChecker, then pixel units.

Step 3: On-screen text and QR codes

Create QR Label Sheets

Separate possible text values with commas. The app will search for a match within this list.

Step 4: Features to measure

Step 5: Manual feature settings

These sliders only apply when Legacy Features (TA) are selected. If they are disabled, click Next to continue.

10%
10%
5%
25%

Analysis setup summary

Review your setup below. Use Edit to change a step, or fine-tune the exact output columns with the Columns menu. You can now schedule jobs from the Image Analysis tab.

Data compatibility check (optional)

Upload 2–5 sample images from your dataset to see how similar they look to the data this model was trained on. Supported file types: JPG, JPEG, and PNG. This check is advisory only and never blocks processing.

Example Image and Output Previews

The first card shows the kind of input image expected by the app. The cards to the right show preview outputs; any additional features you select are returned as numeric or text data columns. Purple preview cards are adjustable after processing: open them to repaint masks or change the visual Legacy Features (TA) controls.

Example fruit image expected by the program
Expected imageThis is the sort of image the program expects: a clear fruit image with the target cut-face visible.
Example on-screen text preview output
Text previewExpected output preview for detected visible text labels; selected text features are also returned as data columns.
Example cleanup mask preview output
Cleanup previewExpected output for cleaned masks. After processing, this preview can be opened to repaint masks, recalculate measurements, and save a fine-tuning correction.
Example smoothing preview output
Smoothing previewExpected output for function fits. Open it after processing to repaint masks or drag weighted curve anchors and refit the rind or flesh smoothing function.
Example traditional feature preview output
Traditional preview (TA)Expected output for Legacy Features (TA). Open it after processing to repaint masks or drag the width, angle-span, and indentation controls.

Setup complete. Run analysis in the Image Analysis tab by uploading one or more images.

Settings are locked for the current batch so resumed images use the same values.

Help

Image Analysis

Upload one or more images. Supported file types: JPG, JPEG, and PNG.

?

Labeling Studio

Pre-label images with the current model, correct the masks with a brush and eraser, then export a YOLO training bundle. Images and labels stay in the platform dataset store.

Select a dataset to see which model will be fine-tuned or copied.
Dev/Production model and dataset sync

Dev controls sync. Production only accepts or provides selected bundles when dev requests them; it never pulls from dev on its own.

Local dev datasets

Local dev models

Production datasets

Production models

PaCMAP Latent Space

Stored expert reference embeddings only. Scroll to zoom, drag to pan, hover points for image/model details.

Augmented export options

These settings generate transformed training copies from approved masks. Preview uses the same augmentation pipeline at a smaller display resolution; export retains the configured training resolution.

Background augmentation mix

The four replacement probabilities form a background mix; any remaining fraction keeps the original background, and totals above 1 are normalized. Ordinary clutter and out-of-context objects are added independently and can coexist with any background. Set a probability to 0 to disable that family.

Fruit-free donor ?
Original background color ?
Blank color ?
Table/counter surface ?
Non-produce clutter ?
Out-of-context objects ?
Augmentation queued.
Training queued.
Elapsed: 00:00:00 | ETA: Calculating...
100%

Select an image to start labeling.