Medicinal plant seed image dataset for attribution-descriptor analysis

This dataset supports the study “Contextualizing CNN attribution maps using predefined image-derived descriptors in medicinal plant seed classification.” It contains 1,124 medicinal plant seed samples across five classes (ARSE, ARSS, PJNA, PRDA, and PRPE): 1,124 original JPEG photographs (4032 × 3024), 1,124 foreground-centered RGB PNG images (512 × 512), and 1,124 binary foreground masks (512 × 512). The release also includes canonical sample metadata, the fixed 899/225 train/test split, the 24-image manual-orientation correction manifest, preprocessing logs, ConvNeXt-Small test predictions, principal result tables, documentation, and SHA-256 checksums. Original JPEG files are preserved byte-for-byte with their EXIF metadata; no GPS metadata were detected. The non-study source image PJNA_0229.jpg was excluded because it was not part of the canonical 1,124-sample dataset or reported analyses. Analysis code is available at https://github.com/sgn08178-bit/medicinal-seed-attribution-descriptors and archived at https://doi.org/10.5281/zenodo.21537321.