Skip to content

Datasets Catalog

Public datasets for cuvis-ai, hosted on Hugging Face under the cubert-gmbh organisation. Each page below mirrors the dataset's Hugging Face README; the HF page is the authoritative source.

Captured 5 with a Cubert Ultris XMR (61 bands, 430–910 nm); 1 with a Cubert Ultris X4 + SWIR rig (three visible and three short-wave-infrared bands).

The table is generated from the registry in cuvis-ai-core (uv run dataset list). uv run dataset download <name> --data-dir data fetches a dataset into data/<folder> and records a .cuvis-dataset.json marker beside it; uv run dataset status --data-dir data reports what is present. In CuvisNEXT the same list lives under Settings › Cuvis.AI › Datasets.

Dataset Camera Size Files Task Licence Download
Demo: blood perfusion XMR 11.2 GB 6 Statistical Apache-2.0 uv run dataset download Blood_Perfusion
Demo: industrial FOD, lentils XMR 6.4 GB 21 Anomaly detection, Segmentation Apache-2.0 uv run dataset download Demo_Industrial_FOD_Lentils
Demo: object tracking XMR 24.4 GB 18 Tracking, Segmentation Apache-2.0 uv run dataset download Demo_Object_Tracking
Industrial FOD, bedding (X4 SWIR) X4 SWIR 178.2 GB 683 Anomaly detection, Segmentation Apache-2.0 uv run dataset download Industrial_FOD_Bedding
Industrial FOD, lentils XMR 57.0 GB 82 Anomaly detection, Segmentation Apache-2.0 uv run dataset download Industrial_FOD_Lentils
Lentils (single session) XMR 921 MB 4 Statistical Apache-2.0 uv run dataset download Lentils

Available datasets

  • XMR_Demo_Blood_Perfusion

    Tissue oxygenation visualised via two-band differential. Pairs with the Blood Perfusion tutorial. No training manifest: a statistical demo.

  • XMR_Demo_Industrial_FOD_Lentils

    Lentil-conveyor scene with stones, stems, aluminium shards, and flies. Pairs with the AdaCLIP tutorial and is the inference showcase of the trained Dinomaly pipelines on its companion model repo. No training manifest: the README declares no split.

  • XMR_Demo_Object_Tracking

    Crowded bus-station scene with passive and active (invisible-ink) tracking modes. No training manifest: the tracking nodes do not train.

  • X4_SWIR_Industrial_FOD_Bedding

    252 VIS + SWIR frames of bedding: 193 normal training frames and a 59-frame validation split holding the 51 annotated foreign-object frames (configs/data/industrial_fod_bedding.yaml).

  • XMR_Industrial_FOD_Lentils

    Fifteen merged conveyor sessions over three days, 1,136 frames (696 annotated), seven foreign-object classes, with baked train / val / test selectors (configs/data/industrial_fod_lentils.yaml, industrial_fod_lentils_normals.yaml); the demo's trained Dinomaly pipelines were fitted on it.

  • XMR_Lentils

    One lentil-conveyor session with COCO annotations, the recording the test suite and configs/data/lentils.yaml use.