The figures and screen recordings with real seeds use public datasets of images and measurements. Each block below says what the dataset is, where it appears on the site, where it came from, its license and how to cite it. The licenses were checked against the original sources on 8 October 2026. When the origin or license of a dataset still needs checking, the block says so. The animations are a different matter: they are drawn in code, and the end of the page explains the difference.
With an image on the site
Datasets that appear in a figure or screen recording
Photos, crops or outlines from these datasets are on the pages listed in each block.
Images and photos from the orchid research group (GPEOrq) and the seed research group (GPSEM)
Scans of orchid seeds after tetrazolium, made on the group's flatbed scanner, and photos of the lab and the team.
Source: GPEOrq and GPSEM. To use these images, write to the contact address on the site.
Photos of the plants and seeds
Photos of orchids, grasses, soybeans, corn, rice, coffee and beans, and of the fun facts, to show each crop whole, from plant to seed. They are by third parties under open licenses, and each one appears with its author and license in the caption.
Sources: Wikimedia Commons, iNaturalist and USDA PLANTS. The photos were downsized and converted to WebP, with no other changes.
Two images from Kew, for comparison
The capture bench and a slide read by both the person and the model, from the Millennium Seed Bank project that trains a model for the tetrazolium test in orchids. They appear only next to our own images, to compare the two approaches.
Citation: Sementes de Orquídeas, version 8. Roboflow Universe. universe.roboflow.com/sementes-de-orqudea/sementes-de-orquideas.
Composite sheet and 145 silhouettes
This is not a dataset but an assembly. Seeds from five datasets with one seed per photo (rice, coffee, corn, soybeans with defects and 88 species) were cropped and pasted into a single scene, with randomized position and rotation. The seeds were not photographed together. The quality-classes figure, with 23 seeds, is a scene of the same kind, using only the soybeans with defects. The 145 silhouettes used in the instruments, in the shape figures and in the Hilum symbol come from the same five sources.
Sources: the Rice, Coffee, Corn, Soybeans with defects and 88 species blocks on this page, all with the license checked. Because coffee is CC BY-SA 4.0, the assembly is released under CC BY-SA 4.0.
Rice, five cultivars
Photos of rice grains of the cultivars Arborio, Basmati, Ipsala, Jasmine and Karacadag, one grain per photo, and a table of 106 shape and color descriptors measured on those photos. We checked SeedCounter's measurements against that table on 1,904 grains.
Citation: Koklu M., Cinar I., Taspinar Y. S. (2021). Classification of rice varieties with deep learning methods. Computers and Electronics in Agriculture 187, 106285. doi:10.1016/j.compag.2021.106285. Also: Cinar I., Koklu M. (2022). Identification of rice varieties using machine learning algorithms. Journal of Agricultural Sciences 28(2), 307–325. doi:10.15832/ankutbd.862482
Soybeans with defects
Photos of soybean seeds, one per photo, in five classes: intact, broken, immature, with a damaged seed coat, and spotted.
Citation: Lin W., Fu Y., Xu P., Liu S., Ma D., Jiang Z., Zang S., Yao H., Su Q. (2023). Soybean image dataset for classification. Data in Brief 48, 109300. doi:10.1016/j.dib.2023.109300. The dataset is at doi:10.17632/v6vzvfszj6.6.
Soybean pods (YOLO POD)
Photos of soybean plants pulled up and laid on black cloth, with a box marked on each pod, made for counting pods. The outlines in the recording were generated inside the boxes and are not part of the original dataset.
Source: the AgML copy on Hugging Face, which assigns the CC BY 4.0 license. The original is on the Google Drive linked in the paper below, with no declared license.
Citation: Xiang S., Wang S., Xu M., Wang W., Liu W. (2023). YOLO POD: a fast and accurate multi-task model for dense Soybean Pod counting. Plant Methods 19, 8. doi:10.1186/s13007-023-00985-4
Three soybean cultivars
Scans of soybean seeds of the cultivars Anjasmoro, Dega 1 and Grobogan, with each seed cropped and the mask in the file itself.
Citation: Syahraza M. A., Hanafiah D. S., Purnamasari F., Nurhasanah R. (2025). Image Dataset of Local Indonesian Soybean Seed Varieties (Anjasmoro, Grobogan, and DEGA-1). Mendeley Data, version 3. doi:10.17632/c733bjz4m3.3
Coffee, four roast levels
Photos of coffee beans, one per photo, at four roast levels: green, light, medium and dark.
Source: the original dataset on Kaggle, under CC BY-SA 4.0. The 224 × 224 copy keeps the same license. Crops and assemblies made from these photos are released under CC BY-SA 4.0.
Citation: Ontoum et al. (2022). Coffee Bean Dataset. arXiv:2206.01841
Corn, three varieties
Photos of corn seeds, one per photo, of the varieties Bhihilifa, SanzalSima and WangDataa.
Source: the images on figshare, under CC BY 4.0, described in the paper below.
Citation: Yuan M., Lv N., Dong Y., Hu X., Lu F., Zhan K., Shen J., Wu X., Zhu L., Xie Y. (2024). A dataset for fine-grained seed recognition. Scientific Data 11, 344. doi:10.1038/s41597-024-03176-5
Only in the calculations
Datasets used only as numbers, with no images
Measurements taken from these datasets appear in tables, charts and in the text. None of their images is reproduced on the site.
Germinating corn, hour by hour
Plates with 10 corn seeds photographed every hour during the germination test, with the state of each seed annotated. The site uses only the annotations, summed per plate: 120 plates, 1,200 seeds.
Citation: Chen C., Bai M., Wang T., Zhang W., Yu H., Pang T., Wu J., Li Z., Wang X. (2024). An RGB image dataset for seed germination prediction and vigor detection: maize. Frontiers in Plant Science 15, 1341335. doi:10.3389/fpls.2024.1341335
Dry beans, seven cultivars
Table of 16 size and shape descriptors measured on 13,611 dry bean grains of seven cultivars.
Citation: Koklu M., Ozkan I. A. (2020). Multiclass classification of dry beans using computer vision and machine learning techniques. Computers and Electronics in Agriculture 174, 105507. doi:10.1016/j.compag.2020.105507
Durum wheat on a conveyor belt
Durum wheat kernels filmed on a conveyor belt, in three classes (vitreous, starchy and foreign matter), with a spreadsheet of descriptors for each object.
Citation: Kaya E., Saritas İ. (2019). Towards a real-time sorting system: identification of vitreous durum wheat kernels using ANN based on their morphological, colour, wavelet and gaborlet features. Computers and Electronics in Agriculture 166, 105016. doi:10.1016/j.compag.2019.105016
Soybean after tetrazolium
Global color and texture descriptors of soybean seed halves scanned after the tetrazolium test, classified by type and intensity of damage. The dataset publishes the descriptors, without the images.
Citation: Pereira D. F., Bugatti P. H., Lopes F. M., Souza A. L. S. M., Saito P. T. M. (2019). Contributing to agriculture by using soybean seed data from the tetrazolium test. Data in Brief 23, 103652. doi:10.1016/j.dib.2018.12.090
Seeds of 65 grass species
Photos of 775 whole seeds of 65 grass species, used to compare shape-only descriptors with shape-and-color descriptors.
Source: origin to be confirmed. The reference will be added here as soon as it has been checked.
The Theory pages also use numbers published in papers, such as the constants of the viability equation. These numbers are cited on the page itself, in the reference list.
Simulation or real data
What is drawn and what is measured
Every figure on the site carries a badge. It says whether the image comes from real seeds or was drawn in code.
Real data
Real seeds
A figure or recording made with one of the datasets on this page. The numbers come from SeedCounter's calculations on that data. The composite sheet, assembled from several photos, is explained above.
Simulation
Animations drawn in code
The animations on the site, on the home page, in Learn, in For labs, in Theory and on the blog, use model seeds drawn in code. They show the idea behind each measurement, not a measured result. The plates in the Test your eye challenge and the 3D bench in For labs are also simulations.
Demo
Examples that come with the app
The Soybean, Orchid TZ and Forage scenes inside SeedCounter are synthetic, with the correct answer known. The screen recordings with the twelve soybean seeds use that demo scene.
Real, coming soon
Dashed frame
Marks where a real photo or recording will go, with the code from the examples list or the recording script.
Is a credit missing, or is one wrong?
If you published one of these datasets and want a different form of citation, or you found an error on this page, write to us. We will correct the page.
