Datasets:

---

cubert-gmbh

/

XMR_Industrial_Foreign_Object_Detection_Lentils

like 0

Follow

Cubert GmbH 3

Adds a universe.csv sample-universe manifest in the shared cuvis-ai-dataloader vocabulary, so the dataset loads directly with the cu3s_multi DataModule (universe_csv=...) together with the shipped splits/*.json selectors.

Columns:

• source: the recording, relative posix path (matches the splits/*.json selector keys and splits.csv cu3s_path).
• index: measurement index within the recording (== COCO image_id).
• annotation: the per-recording COCO json (relative), for pixel labels.
• group: leakage-grouping key (carried from splits.csv group_id).

1136 rows (every frame in splits.csv). No split column: the universe is selector-driven, so splits/dinomaly.json (636 frames) and splits/adaclip.json (1136 frames) pick the subsets; both resolve fully against this universe (0 missing). Paths are relative to the dataset root, so the manifest is portable.