Real seeds, step by step.

Morphometry · checked against the literature

Rice: 7 of 11 measurements match the published ones

On 1,904 rice grains, SeedCounter computed the same shape descriptors as a published dataset (Koklu et al.). Seven matched with an error of 0.00%. The remaining differences come from calculation conventions, and they are explained.

1. Area, perimeter, major and minor axes, eccentricity, solidity and circularity.
2. The shapes separate in shape space, with no one saying which is which.

In preparation: two types of rice photographed with a ruler on the bench, from the click to shape space.

Quality classes · public dataset

Soybean: what is each seed?

Intact, broken, immature, with a damaged seed coat, or spotted. Counting is the first step; the question for anyone buying seed is how much of the seed lot falls in each class. The colors in the figure come from the annotation of the source dataset.

1. Load the image of the seed lot with the ruler.
2. Create the classes for your protocol, each with a color and a marker shape.
3. Outline each seed and assign its class, checking in the gallery.
4. Export the count by class with percentages.

Counting · multiple crops

Rice, corn and coffee on the same sheet

Different shapes, colors and sizes in the same image. Each seed found gets a number, and the total count sits in the corner, ready to check.

1. Load the image and calibrate the scale.
2. Outline the seeds yourself or let the optional AI suggest them.
3. Show the indices to number each seed on screen.
4. To project it in class, open presentation mode, with a clean interface and the marks on the image.

Morphometry · gallery and dataset hub · version 4.0

Seed size across the lot, one seed at a time

After outlining, the gallery stacks the cropped seeds by length bin, with the ruler underneath. The mean is shown, but the picture reveals what the mean hides: whether the seed lot is uniform, whether it has two populations, whether some seeds are atypical.

In the dataset hub, the same histogram comes from a whole dataset: each polygon in the labels becomes a cropped seed, carrying the label's class. In the screen recording, 20 tetrazolium photos of orchid seeds become 917 seeds, and the hub steps through its five views: the histogram by class, the overlaid densities, the mosaic lined up by the chosen measurement, the mean shape with the shape space, and the color in a* × b*, where the viable seeds separate out by their tetrazolium red. In length, the viable seeds come out longer than the non-viable ones (135 vs. 100 px on average). Part of that is zoom: without a ruler, the pixel measures the seed and the framing together, and within a single crop the difference drops to about 5% (the calculation is in Shape).

1. Outline the seeds on the plate.
2. Open the gallery and turn on Histogram.
3. Change the measurement (length, width, area) and the number of bins.
4. Use the Outside the mean filter to review the seeds in the tails.
5. For a whole dataset, open the hub, draw a random sample per class and build the histogram; then switch the view.

Counting and measuring · dataset hub · public dataset

Soybean pods, one by one

Soybean plants pulled up and laid on black cloth, from a public dataset built for counting pods. The dataset labels give one box per pod; the outlines were generated automatically inside each box and go into the hub as polygons. From 8 randomly drawn photos, the hub crops 392 pods and shows the length distribution, the pods lined up by the chosen measurement and the mean shape.

1. Open the dataset's folder in the dataset explorer and choose Hub.
2. Pick photos at random and build the histogram.
3. Switch to Density, Mosaic and Shape.
4. Check the pods at the ends in the Mosaic: that is where the bad outlines show up.

Morphometry · dataset hub · public dataset

Three soybean cultivars side by side

Soybean seeds of the cultivars Anjasmoro, Dega 1 and Grobogan, scanned and cropped one by one by researchers in Indonesia, with each seed's mask in the file itself. There are 40 seeds of each cultivar, one row per cultivar in the histogram. The scans in the dataset range from 800 to 950 dpi, so the size difference in pixels between cultivars only turns into millimeters after calibration.

1. Open the dataset in the hub and build the histogram with the 120 seeds.
2. Compare the cultivars in the overlaid densities.
3. In the Shape view, look at each cultivar's mean seed shape: almost the same ellipse.
4. In the Mosaic, the seeds line up in order of length, from smallest to largest.

Counting and measuring

Soybean: count and measure the seed lot

A commercial soybean sample on a dark background, with a ruler in the same plane. The starter case: large, separated seeds, to see the whole workflow and how long each step takes.

1. Load the photo and mark 10 mm on the ruler to set the scale.
2. One click per seed with the wave tool, or the optional AI proposing all of them.
3. Check in the gallery, seed by seed.
4. Export the spreadsheet with area, length and width in millimeters.

Tetrazolium

Orchid: color decides

Seeds of about 1 mm, elongated and often touching. In tetrazolium, the viable embryo turns red. The analyst marks each seed viable or non-viable, with a disc and a ring, and the plate's viability is calculated on the spot.

1. Load the photo of the plate taken under the stereo microscope, with the scale.
2. Outline the seeds and let the app propose viable or non-viable.
3. Review the proposal in the gallery, filtering by class.
4. Read the plate's viability in the panel, along with the number of seeds checked.

Protocol classes

Forage grass: full or empty

Spikelets that look alike on the outside. The full one has a caryopsis and the empty one does not. This is the case where viable and non-viable are not enough and the protocol classes come in: "full", "empty", "dormant", "dead".

1. In the forage grass profile, the protocol classes are right at hand.
2. Mark each spikelet and check the proportion of each class.
3. Save the session to resume or compare seed lots later.

Germination · over the days

Germination: from the daily count to t50

Each count in the test becomes a point. The Germination tab builds the cumulative curve, calculates indices such as t50 and compares treatments. To track a stored seed lot, you repeat the same test each time seed is taken out of storage.

1. Photograph the same gerbox (germination box) every day, with the same framing.
2. Record the counts for each day of the test.
3. See the cumulative curve and the t50 for each treatment.
4. Compare seed lots and export the data for your own statistics.