maxwildersmith/splataverse-sketchfab

Splataverse: Sketchfab

Removed objects (quality filter v1, 2026-10-05)

1,525 objects were removed after an automatic quality survey of their splats: untextured flat-grey objects (no colour information), and objects whose splats were mostly disconnected fragments, far-off floaters or oversized blurry Gaussians. Flat but strongly coloured objects were kept. removed_quality_v1.csv lists every removed uid with the reason, the metric values and the paths it had.

Removed objects (quality filter v2, 2026-10-05)

The same filter, scored directly on the splats, was then run over every Sketchfab object in the frozen split (data/splits/v1.parquet): 8,685 more objects were removed, including 598 empty or non-finite splats. removed_quality_v2.csv lists them with reasons and former paths.

Splits and quality flags (v3)

Every object was scored on its splat during the SPZ v4 re-encode (colour spread and saturation, share of Gaussians outside the main piece, bbox outlier ratio, share of oversized splats, empty splats). quality/quality_flags.parquet has every scored file with its metrics and reasons ("" = clean). Flagged objects are kept in the repo, only excluded from the training tiers.

splits/v3.parquet (config splits_v3) gives each object one split: train 459,890, flagged 266,962, unscored 5,243, test 1,000, val 1,000. Some objects are stored twice under different shard folders; the manifest lists one path per object. Training tiers are nested (1k is inside 5k is inside 10k ...); tier T is split == "train" & tier <= T. val and test are fixed, clean, licence-stratified and disjoint from every tier.

path list	objects
val	1,000
test	1,000
train_1k	1,000
train_5k	5,000
train_10k	10,000
train_25k	25,000
train_50k	50,000
train_100k	100,000
train_250k	250,000
train_all	459,890

from huggingface_hub import hf_hub_download, snapshot_download

repo = "maxwildersmith/splataverse-sketchfab"
paths = open(hf_hub_download(repo, "splits/sketchfab/train_5k.txt", repo_type="dataset")).read().split()
snapshot_download(repo, repo_type="dataset", allow_patterns=paths, local_dir="splataverse_sketchfab_5k")