Luxar demo datasets: permissively licensed (CC-BY, CC0, public domain)

Precomputed demo datasets for Luxar, a system for compiling and visualising arbitrary-sized nD scientific scenes in the browser. These files are expensive to compute (GPU fitting, in some cases hours) and are hosted here so the software repository stays small and installable. Every file is a derived product: a Gaussian-splat fit, a decimated level-of-detail structure, or an extracted point catalog. None of them is the raw source imagery. The upstream sources are cited per dataset below, and users of this record should cite the upstream source in addition to this record. Licensing This record groups datasets whose upstream licences permit redistribution with attribution: CC BY 4.0, CC0 1.0, and United States federal public domain. The record carries a single CC BY 4.0 stamp because Zenodo permits one licence per record, and CC BY is the most restrictive of that set; the effective licence of an individual file is the one listed against it below. Datasets under ShareAlike terms are kept in a separate record, because ShareAlike survives derivation and would otherwise contaminate this one. How to use these files Each file is a .gsplats.zarr.zip or .zarr.zip archive that the Luxar demo suite fetches on first run and caches under ~/.cache/luxar/ /. To fetch one directly: from luxar.demos import ensure_dataset paths = ensure_dataset("gsplats_kidney") # downloads, verifies, returns paths To open one in the viewer without writing code: luxar gsplat view .gsplats.zarr.zip luxar demo # browse every demo that uses these files Every file is pinned by SHA-256 in data_manifest.json, and the fetch helper verifies the checksum before use; a mismatched download is quarantined rather than returned. How the Gaussian-splat files were produced Each source volume was fitted with the Luxar gsplat pipeline (luxar gsplat fit) on a GPU. These are lossy representations built for visualisation and interactive exploration; they are not a substitute for the source imagery in quantitative analysis. Choosing the splat count. The number of Gaussians K was calibrated per dataset by Noise2Self blind-spot cross-validation (luxar gsplat cal), and then checked by refitting at the candidate count and scoring the reconstruction against the whole source volume. The second step is not a formality. The cross-validated estimate is computed over a five percent mask of held-out voxels, and on these datasets it disagreed with the direct measurement in both directions: for one dataset it favoured half the splat budget that measurement showed to be better, and for another it predicted a gain of about 1.5 dB where refitting delivered under 0.3 dB. Where two candidate budgets fell within roughly a decibel of each other, the choice was made from renders of both at matched magnification, because at that margin the metric and the eye disagree. Additional Gaussians capture real structure and photon noise alike, and peak signal-to-noise ratio rewards the two equally. Level of detail These files carry level-of-detail structure so a viewer can show something immediately and refine it as the camera approaches, instead of waiting for a whole fit to arrive. Two independent mechanisms are used, and a file may carry both: Progressive (additive) ladders. A leaf is stored as an ordered series of chunks whose union is the complete fit, so any prefix is a valid, dimmer rendering of the whole object. Every file in this record carries one. Substitutive levels. The Milky Way dust volume and the seven cell-tracking timelapses additionally carry coarse levels of merged representative Gaussians, at one quarter, one sixteenth and one sixty-fourth of the full count, each of which replaces the level below rather than adding to it. The viewer selects a level from the fraction of the screen the object occupies, so the choice is independent of display resolution. The whole-slide pathology images use a third arrangement: spatial tiles, each carrying its own coarse and fine levels, so that panning and zooming over a gigapixel slide loads only the tiles in view at only the detail required. Reconstruction quality Two figures are reported per file, and both are stamped inside the archive itself rather than only quoted here: PSNR over the whole volume, in decibels. Foreground PSNR, over only those voxels above an Otsu threshold of the source volume, with the fraction of voxels that threshold selects. The foreground figure is the honest one for sparse microscopy and astronomy data. Where a volume is mostly empty, a global PSNR is dominated by correctly reconstructed emptiness and can look excellent while the structure of interest is poorly represented. A low foreground figure is also not automatically a defect: on a noisy channel it partly measures photon noise that a sum of Gaussians declines to reproduce, which is desirable behaviour that this metric charges for. Where that applies it is noted against the dataset. To read the stamps from a downloaded file, so that nothing here has to be taken on trust: luxar gsplat info .gsplats.zarr.zip Each archive also records the grid and element type of the volume it was fitted from, which is what makes the compression ratio quoted against it reproducible rather than asserted. Contents file size elements level of detail PSNR / foreground (dB) 3d_umap_coords_human.parquet 35.5 MB 1,041,455 table n/a (coordinates) 3d_umap_coords_mouse.parquet 7.3 MB 192,251 table n/a (coordinates) 6bba_09961292.gsplats.zarr.zip 91.9 MB 4,054,345 4 coarse-to-fine levels, 100 timepoints 24.7–26.5 / 18.3–20.2 (per frame) 6bba_09961292_tracks.npz 0.03 MB 1,950 lineage tracks (points + polylines) n/a (sidecar) 6bba_57b7cc1e.gsplats.zarr.zip 51.0 MB 2,178,516 4 coarse-to-fine levels, 100 timepoints not scored 6bba_57b7cc1e_tracks.npz 0.03 MB 1,659 lineage tracks (points + polylines) n/a (sidecar) 6bba_5dfe9ad1.gsplats.zarr.zip 105.0 MB 4,649,701 4 coarse-to-fine levels, 100 timepoints not scored 6bba_5dfe9ad1_tracks.npz 0.03 MB 1,542 lineage tracks (points + polylines) n/a (sidecar) 6bba_784a78c9.gsplats.zarr.zip 101.6 MB 4,520,586 4 coarse-to-fine levels, 100 timepoints not scored 6bba_784a78c9_tracks.npz 0.03 MB 1,788 lineage tracks (points + polylines) n/a (sidecar) 6bba_ae82a791.gsplats.zarr.zip 93.6 MB 4,197,409 4 coarse-to-fine levels, 100 timepoints not scored 6bba_ae82a791_tracks.npz 0.02 MB 1,495 lineage tracks (points + polylines) n/a (sidecar) 6bba_bb9f20c3.gsplats.zarr.zip 96.3 MB 4,283,665 4 coarse-to-fine levels, 100 timepoints not scored 6bba_bb9f20c3_tracks.npz 0.03 MB 1,925 lineage tracks (points + polylines) n/a (sidecar) 6bba_cff5865f.gsplats.zarr.zip 92.8 MB 4,101,691 4 coarse-to-fine levels, 100 timepoints not scored 6bba_cff5865f_tracks.npz 0.03 MB 1,571 lineage tracks (points + polylines) n/a (sidecar) celegans_s1.gsplats.zarr.zip 80.8 MB 5,640,070 400 per-frame stores 29.5–39.8 / 21.8–38.1 (per frame) cells3d_ch0.gsplats.zarr.zip 0.5 MB 41,975 4 ladder rungs 32.8 / 32.0 cells3d_ch1.gsplats.zarr.zip 0.8 MB 57,946 4 ladder rungs 22.4 / 21.7 census_umap_1m.npz 13.1 MB 1,000,000 table n/a (coordinates) cmu1_ch0.gsplats.zarr.zip 45.7 MB 8,823,953 4 parts, 16 ladder rungs 23.5 / 16.6 cmu1_ch1.gsplats.zarr.zip 50.7 MB 9,924,486 4 parts, 16 ladder rungs 24.3 / 18.7 cmu1_ch2.gsplats.zarr.zip 53.7 MB 10,830,790 4 parts, 16 ladder rungs 26.1 / 22.3 cryoem_virus.gsplats.zarr.zip 10.6 MB 1,016,482 flat, 5 ladder rungs 28.2 / 19.2 ct_atlas.gsplats.zarr.zip 7.0 MB 647,083 flat, 4 ladder rungs 43.0 / 30.6 ct_atlas_labels.npz 0.1 MB 647,083 per-splat organ ids, also carried inside the fit n/a (sidecar) dapi.gsplats.zarr.zip 0.1 MB 7,740 4 ladder rungs 43.8 / 34.8 desi_dr1_cosmic_web.luxar.zarr.zip 76.8 MB 9,751,955 2 layers x 4 levels n/a (not a fit) droso_gastrulation.gsplats.zarr.zip 2.2 MB 200,023 3 ladder rungs 41.6 / 37.1 flylight_mcfo_63x.gsplats.zarr.zip 7.7 MB 660,035 5 ladder rungs 60.9 / 38.9 h2afva_full.gsplats.zarr.zip 17.8 MB 1,653,405 single leaf 63.7 / 48.9 h2afva_quarter.gsplats.zarr.zip 4.6 MB 411,180 single leaf 57.1 / 41.7 h2afva_tenth.gsplats.zarr.zip 1.9 MB 165,276 single leaf 53.8 / 38.3 h2afva_tp234.gsplats.zarr.zip 20.8 MB 1,653,405 41 parts, 246 ladder rungs 63.7 / 48.9 h2afva_twentieth.gsplats.zarr.zip 1.0 MB 82,660 single leaf 51.7 / 36.2 kidney_ch0.gsplats.zarr.zip 0.7 MB 51,015 4 ladder rungs 25.8 / 24.1 kidney_ch1.gsplats.zarr.zip 0.7 MB 52,278 4 ladder rungs 35.6 / 30.6 kidney_ch2.gsplats.zarr.zip 0.7 MB 53,827 4 ladder rungs 32.0 / 26.3 milkyway_dust.gsplats.zarr.zip 10.6 MB 677,741 4 coarse-to-fine levels, each progressively streamed 34.8 / 21.1 neuromast_membranes.gsplats.zarr.zip 72.4 MB 6,400,000 8 ladder rungs not scored neuromast_nuclei.gsplats.zarr.zip 70.4 MB 6,400,000 8 ladder rungs not scored nexrad_supercell.gsplats.zarr.zip 12.9 MB 817,989 82 per-frame stores 43.3–56.8 / 32.0–39.7 (per frame) blastocyst_ch0.gsplats.zarr.zip 0.2 MB 12,916 4 ladder rungs 38.6 / 27.0 blastocyst_ch1.gsplats.zarr.zip 0.2 MB 13,941 4 ladder rungs 38.6 / 28.9 vh_head.gsplats.zarr.zip 25.1 MB 1,908,888 flat, 4 ladder rungs 38.2 / 32.6 vh_head_colors.npz 5.1 MB 1,908,888 per-splat RGB, also carried inside the fit n/a (sidecar) Every figure above was measured from the file this record holds. “Elements” is what a viewer draws at full detail: splats summed over the finest level of each part, or rows for a coordinate table. Foreground PSNR is restricted to signal voxels and is the number to judge a fit by — a global PSNR on a mostly-empty volume largely measures how well the fit reproduces empty space. A timelapse bundle is scored per frame and quoted as the RANGE across frames rather than as one number for the whole series, because quality genuinely varies over a recording — a fixed per-frame splat budget covers a subject that grows. For the per-frame bundles that range spans every frame in the archive; for t