musharna/lobelia-silhouettes
Lobelia silhouettes 39 whole-plant silhouettes across 27 species of Lobelia, digitally restored from herbarium specimens between April 2018 and February 2019. Every silhouette carries true physical scale. A pressed herbarium specimen is a plant that has been folded to fit a sheet, mounted with tape and glue, annotated, and photographed decades later. Each sheet here was restored in Photoshop — leaves digitally unfolded, snapped stems rejoined, labels, tape, mounting strips and… See the full description on the dataset page: https://huggingface.co/datasets/musharna/lobelia-silhouettes.
Lobelia silhouettes
39 whole-plant silhouettes across 27 species of _Lobelia_, digitally restored from herbarium specimens between April 2018 and February 2019. Every silhouette carries true physical scale.
A pressed herbarium specimen is a plant that has been folded to fit a sheet, mounted with tape and glue, annotated, and photographed decades later. Each sheet here was restored in Photoshop — leaves digitally unfolded, snapped stems rejoined, labels, tape, mounting strips and colour cards removed — and the restored plant converted to a mask. That mask is the silhouette.
Why these are restorations, not illustrations
- Nothing is added that was not on the sheet. No compositing across specimens, no reconstructing an organ from a better example elsewhere, no idealised species representative. If a part was missing from the physical sheet, it is missing from its silhouette.
- Scale comes from the sheet's own ruler. Every specimen was photographed with a scale bar, and each silhouette carries a 10 cm bar drawn into its own file. The set is comparable in true size, and that claim is checkable from the images rather than taken on trust.
A standard herbarium sheet is about 42 cm tall, yet L. brevifolia restores to 133.8 cm and L. apalachicolensis to 98 cm. A plant taller than the sheet it is mounted on is exactly what unfolding recovers.
Fields
Heights run from L. feayana at 13.6 cm to L. brevifolia at 133.8 cm. Seven species have more than one restored specimen (four L. anatina; three each of L. appendiculata, L. cardinalis and L. nuttallii), so within-species spread is visible — the three L. cardinalis restore to 24, 47 and 52 cm.
⚠️ Read this before rescaling
Files are not stored at a common resolution. px_per_cm ranges from 30.4 to 85.1 across the set. To compare shapes at true size you must rescale using px_per_cm, the 10 cm bar in each file, or truescale_plate.png (which was built that way).
Do not derive scale from `width_cm`. width_cm is the plant's width, but image width is set by whichever is wider — the plant or the drawn-in 10 cm scale bar. For narrow specimens the bar wins, so px_width / width_cm overestimates scale by up to 81 % (worst case L. feayana). Only the height-derived factor is valid, which is why px_per_cm is precomputed here.
The px_per_cm column was verified against the drawn scale bars themselves: measuring bar length in pixels directly reproduces the manifest heights to within 0.1 cm, and the resulting 30.4–85.1 range matches the range reported independently in the original deposit.
Provenance and rights
The silhouettes are original work by the author, released under CC BY 4.0.
They derive from digitized herbarium sheets held by the institutions listed in source_sheets.csv. What survives into a silhouette is the outline of the specimen — a fact about the plant — together with the author's restoration work; none of the source photographs' pixels, tonality, lighting or framing are reproduced. The source images themselves are not redistributed here and remain under their institutions' own terms.
source_sheets.csv covers six species only (cardinalis, canbyi, elongata, dortmanna, brevifolia, feayana), not all 27. It is a species-level record of provenance, not a per-silhouette mapping — it does not assert which sheet became which file. Institution and licence columns were resolved against GBIF in August 2026; 46 of 60 accessions matched confidently and the rest are recorded as unresolved rather than guessed. A licence field on an aggregator record is the aggregator's assertion, not necessarily the rights holder's.
Also not included: no specimen photographs, and no layered working files (the per-specimen Photoshop documents embed the source photographs directly).
Provenance of the work itself
Undergraduate research at Kent State University, Department of Biological Sciences, in the lab of Dr. Andrea Case. The artwork predates the grant it is usually associated with: the silhouettes were finished by early 2019, while NSF award DEB-2015606 began in September 2020.
The set is credited in the acknowledgements of Godden et al. (2025), where the silhouettes form the morphology panel of the phylogeny figure. That is an acknowledgement of artistic contribution, not authorship of the paper.
Godden, G.T., et al. (2025). Population-level phylogenomic analysis yields insights into species cohesion and population substructure of Lobelia. Molecular Phylogenetics and Evolution, 108410. <https://doi.org/10.1016/j.ympev.2025.108410>
Citation
@dataset{arnold_lobelia_silhouettes,
author = {Arnold, Jaret},
title = {Lobelia silhouettes: digitally restored whole-plant outlines from herbarium specimens (2018--2019)},
year = {2026},
publisher = {Zenodo},
doi = {10.5281/zenodo.21764521},
url = {https://doi.org/10.5281/zenodo.21764521}
}Links
- Archival deposit (DOI of record): <https://doi.org/10.5281/zenodo.21764521>
- Write-up with figures: <https://musharna.github.io/projects/LobeliaSilhouettes/>
- Related work — semi-automated leaf measurement from dismembered sheets: <https://github.com/musharna/lobelia-leaf-morphometrics>
Two distinct projects, easily confused, because both reduce a pressed plant to an outline. This dataset is the whole-plant silhouettes (2018–19), which preserve true size. The leaf-morphometrics repo is a later, separate pipeline (2021–24) that normalises size away to compare leaf shape only.