{"success":true,"data":{"author":"Mark Schuiveling et al. (PUMA challenge); curated by SPORA","bucketAccessType":"public","bucketName":"csmp-hub/puma","bucketPrefix":"","category":"Health & Biomedical","cloudProvider":"huggingface","createdAt":"2026-11-17T19:29:32.938Z","description":"PUMA (Panoptic segmentation of nUclei and tissue in MelAnoma) is the public training set of the PUMA challenge: 206 H&E regions of interest from melanoma, half from primary tumors and half from metastases, each 1024x1024 pixels. Nucleus instances carry ten classes, made with model assistance and corrected by hand, and tissue regions in six classes were hand-drawn and reviewed by dermatopathologists. puma.parquet holds one row per image with RGB image and masks as PNG bytes; the original GeoJSON nucleus and tissue annotation archives are included alongside. Patient IDs are not available, so image names alone cannot group images by patient, and the challenge's private test images are not included. About 430 MB (measured 2026-11-17). Images and masks are stored as PNG or TIFF bytes inside Parquet; read them with polars or pandas plus Pillow, or with Hugging Face datasets. Use for panoptic nucleus segmentation, TIL quantification and tissue segmentation in skin cancer. Curated and packaged on Hugging Face by the SPORA project (csmp-hub), which keeps the original license and citation files. The authors' Zenodo record 15050523 is released under CC0 1.0; cite Schuiveling et al., GigaScience 2025 (doi 10.1093/gigascience/giaf011).","displayTitle":"PUMA: melanoma H&E images with nucleus instances, ten nucleus classes and six tissue classes","doi":"10.5281/zenodo.15050523","huggingFaceUrl":"https://huggingface.co/datasets/csmp-hub/puma","id":"579f824c-0010-52a4-96ec-0d3fa8616527","license":"CC0-1.0","licenseEvidenceUrl":"https://zenodo.org/records/15050523","objectCountEstimate":9,"probeStatus":"ok","referenceUrl":"https://doi.org/10.1093/gigascience/giaf011","region":"us","schemaCard":{"accessHints":[{"language":"python-datasets","snippet":"from datasets import load_dataset\nds = load_dataset(\"csmp-hub/puma\")"},{"language":"python-hub","snippet":"from huggingface_hub import snapshot_download\nsnapshot_download(repo_id=\"csmp-hub/puma\", repo_type=\"dataset\")"}],"caveats":["HuggingFace metadata: author=csmp-hub, license:cc0-1.0, size_categories:n<1K, format:parquet, downloads=240"],"format":"mixed","generatedAt":"2026-11-17T19:40:08.500Z","layout":{"example":"puma.parquet","pattern":"{repoId}/{filename}"},"mountPath":"/mnt/archive","sampleHeaders":[{"bytes":"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","encoding":"hex","key":"puma.parquet","truncatedAt":512}],"sampleObjects":["01_training_dataset_geojson_nuclei.zip","01_training_dataset_geojson_tissue.zip","examples.png","masks.png","puma.parquet"],"sidecars":[{"bytes":"---\nlicense: cc0-1.0\ntask_categories:\n  - image-segmentation\ntags:\n  - histology\n  - melanoma\n  - nuclei\n  - panoptic\n  - instance-segmentation\nconfigs:\n  - config_name: default\n    data_files:\n      - split: train\n        path: puma.parquet\n---\n\n# PUMA\n\n## Overview\n\nPUMA contains **206 H&E images of melanoma tissue**, with masks for individual nuclei, ten nucleus classes and six tissue classes. Half the images come from primary melanomas and half from metastatic melanomas. Each image is 1024 × 1024 pixels.\n\nNucleus labels were made with model assistance and corrected by hand. Tissue regions were drawn by hand and reviewed by dermatopathologists. The data can be used to study cells and their surrounding tissue in melanoma.\n\n**License:** [CC0 1.0 Universal](https://creativecommons.org/publicdomain/zero/1.0/).\n\n## Files and splits\n\n| Split | File | Images |\n| --- | --- | --- |\n| `train` | `puma.parquet` | 206 |\n\nThis is the public PUMA challenge training set. It has no separate validation or test split; the challenge's private evaluation images are not included. Patient IDs are unavailable, so image numbers alone cannot tell you whether two images come from the same patient. This matters when making your own training and test groups.\n\n## Fields\n\nEach row contains one image and its masks, stored as PNG bytes. Images are 1024 × 1024 pixels with three colour channels; masks are 1024 × 1024 arrays.\n\n| Field | Contents |\n| --- | --- |\n| `id` | Original image name, such as `training_set_metastatic_roi_001` |\n| `image` | RGB colour image; uint8 values from 0 to 255 |\n| `inst` | Numbered nucleus mask; uint16, with 0 for background |\n| `type` | Nucleus class mask; uint8, with 0 for background |\n| `sem` | Tissue class mask; uint8, with 0 for unannotated pixels |\n| `specimen_type` | `primary` or `metastatic` |\n| `mpp_x`, `mpp_y` | Physical pixel size in micrometres along each image axis |\n\nNucleus numbers follow the original GeoJSON annotation order, starting at 1, and can have gaps. An annotation with several separate pieces keeps one number. `type` gives the class of each marked nucleus. These numbers are local to each image.\n\nImage size and colour values are unchanged. Pixel sizes are approximately 0.2262–0.2274 µm; use `mpp_x` and `mpp_y` for measurements at the source scale.\n\n## Labels\n\n### Nucleus classes (`type`)\n\nThe original nucleus and tissue class names use the following integer labels.\n\n| Value | Original class name |\n| --- | --- |\n| 0 | background (no nucleus) |\n| 1 | nuclei_tumor |\n| 2 | nuclei_stroma |\n| 3 | nuclei_endothelium |\n| 4 | nuclei_histiocyte |\n| 5 | nuclei_melanophage |\n| 6 | nuclei_lymphocyte |\n| 7 | nuclei_plasma_cell |\n| 8 | nuclei_neutrophil |\n| 9 | nuclei_apoptosis |\n| 10 | nuclei_epithelium |\n\n### Tissue classes (`sem`)\n\n| Value | Original class name |\n| --- | --- |\n| 0 | unannotated; ignore when training or scoring tissue masks |\n| 1 | tissue_tumor |\n| 2 | tissue_stroma |\n| 3 | tissue_epidermis |\n| 4 | tissue_necrosis |\n| 5 | tissue_blood_vessel |\n| 6 | tissue_white_background |\n\nWhite background is class 6. Unannotated tissue pixels are 0; leave them out when training or scoring tissue masks. In the nucleus masks, 0 means background.\n\n## Download and read a sample\n\n```python\nfrom io import BytesIO\n\nimport numpy as np\nimport polars as pl\nfrom huggingface_hub import hf_hub_download\nfrom PIL import Image\n\n\ndef read_png(data: bytes) -> np.ndarray:\n    \"\"\"Decode RGB pixels or integer labels without reducing mask bit depth.\"\"\"\n    with Image.open(BytesIO(data)) as image:\n        return np.array(image)\n\n\npath = hf_hub_download(\n    repo_id=\"csmp-hub/puma\",\n    repo_type=\"dataset\",\n    filename=\"puma.parquet\",\n)\nrow = pl.read_parquet(path, n_rows=1).row(0, named=True)\nimage = read_png(row[\"image\"])\ninstances = read_png(row[\"inst\"])\ncell_types = read_png(row[\"type\"])\ntissue_types = read_png(row[\"sem\"])\n```\n\n## Plot examples\n\nContinue from the loading example.\n\n```python\nimport matplotlib.pyplot as plt\nfrom matplotlib.patches import Patch\nfrom skimage.segmentation import find_boundaries\n\nsample_ids = [\n ","encoding":"utf8","key":"README.md","name":"README.md","size":10721,"truncated":true},{"bytes":"{\n  \"source\": \"https://zenodo.org/records/15050523\",\n  \"version\": 5,\n  \"license\": \"cc0-1.0\",\n  \"nuclei_classes\": {\n    \"background\": 0,\n    \"nuclei_tumor\": 1,\n    \"nuclei_stroma\": 2,\n    \"nuclei_endothelium\": 3,\n    \"nuclei_histiocyte\": 4,\n    \"nuclei_melanophage\": 5,\n    \"nuclei_lymphocyte\": 6,\n    \"nuclei_plasma_cell\": 7,\n    \"nuclei_neutrophil\": 8,\n    \"nuclei_apoptosis\": 9,\n    \"nuclei_epithelium\": 10\n  },\n  \"tissue_classes\": {\n    \"unannotated\": 0,\n    \"tissue_tumor\": 1,\n    \"tissue_stroma\": 2,\n    \"tissue_epidermis\": 3,\n    \"tissue_necrosis\": 4,\n    \"tissue_blood_vessel\": 5,\n    \"tissue_white_background\": 6\n  },\n  \"rasterization\": \"GDAL pixel-center fill, identity XY coordinates, all_touched=False; later source features win overlaps; no geometry repair. Instance ID is the 1-based source feature index; unrepresented polygons are listed per sample and retained in original GeoJSON archives.\",\n  \"sources\": [\n    {\n      \"file\": \"01_training_dataset_tif_ROIs.zip\",\n      \"md5\": \"1e16d440f0f94156fd5c0ea3f082bd90\",\n      \"sha256\": \"af48b879f8ff7e74b84a7114924881606f13f108aa0f9bcc21d3593b717ee022\",\n      \"bytes\": 646953540\n    },\n    {\n      \"file\": \"01_training_dataset_geojson_nuclei.zip\",\n      \"md5\": \"1f695c66db9461251c581e850f73c044\",\n      \"sha256\": \"eda271225900d6de0759e0281f3731a570e09f2adab58bd36425b9d2dfad91a0\",\n      \"bytes\": 15975301\n    },\n    {\n      \"file\": \"01_training_dataset_geojson_tissue.zip\",\n      \"md5\": \"d9a1d94f078068ed26687cca28b24baf\",\n      \"sha256\": \"fc2835135cc28324f52eac131327f0f12c554c0b1f334a108bf4b65e0f18c42b\",\n      \"bytes\": 3070756\n    }\n  ],\n  \"conversion_script_sha256\": \"555d553770065bf71b86b22f74d0e1fb1367933a1c1aa13e52608151c5ce890b\",\n  \"software\": {\n    \"numpy\": \"2.5.4\",\n    \"pyarrow\": \"23.0.1\",\n    \"rasterio\": \"1.4.4\",\n    \"gdal\": \"3.10.3\",\n    \"shapely\": \"2.2.0\",\n    \"tifffile\": \"2026.9.20\"\n  },\n  \"parquet_sha256\": \"68b4e1809bed1a81e953469b0ede6698e438c3153b118e044602b3588d1b9c73\",\n  \"samples\": [\n    {\n      \"nuclei\": 633,\n      \"raster_nuclei\": 633,\n      \"unrepresented_nuclei\": [],\n      \"nuclei_by_class\": {\n        \"nuclei_tumor\": 502,\n        \"nuclei_apoptosis\": 120,\n        \"nuclei_lymphocyte\": 8,\n        \"nuclei_endothelium\": 3\n      },\n      \"nucleus_overlap_pixels\": 89,\n      \"tissue_overlap_pixels\": 0,\n      \"unannotated_tissue_pixels\": 0,\n      \"invalid_source_geometries\": 0,\n      \"independent_pixel_checks\": 128,\n      \"id\": \"training_set_metastatic_roi_001\",\n      \"mpp_x\": 0.22631602770108178,\n      \"mpp_y\": 0.22632114971144054,\n      \"pixel_sha256\": {\n        \"image\": \"d77cb107aadee575158ef46eb51d0ce805a6706c99309a6a544b2e5f4892a91a\",\n        \"inst\": \"a70b820a1d88493d19a72b3042fff3ece7d1914c2b8a01ceb5ad6f639539a6e3\",\n        \"type\": \"f0b90104e82206f69a6fa209a512ca3b07849caea95338bc18b7153722bad970\",\n        \"sem\": \"779aad1f07e39f8284c8fe461fd97caf4c1936e9a025121ca41ad4741bb1e05a\"\n      }\n    },\n    {\n      \"nuclei\": 824,\n      \"raster_nuclei\": 824,\n      \"unrepresented_nuclei\": [],\n      \"nuclei_by_class\": {\n        \"nuclei_lymphocyte\": 134,\n        \"nuclei_tumor\": 671,\n        \"nuclei_plasma_cell\": 7,\n        \"nuclei_stroma\": 9,\n        \"nuclei_endothelium\": 3\n      },\n      \"nucleus_overlap_pixels\": 609,\n      \"tissue_overlap_pixels\": 0,\n      \"unannotated_tissue_pixels\": 0,\n      \"invalid_source_geometries\": 0,\n      \"independent_pixel_checks\": 128,\n      \"id\": \"training_set_metastatic_roi_002\",\n      \"mpp_x\": 0.2268911376321641,\n      \"mpp_y\": 0.22690658255996007,\n      \"pixel_sha256\": {\n        \"image\": \"8e6293292d65b7d22ebcac441a5fdd20682de420bb780e68d6811666bd22b1d9\",\n        \"inst\": \"b3f45c63622668d57471d7586b3410ca9f04f68d0d623165f99ef7b642ab6cb9\",\n        \"type\": \"19fdd7ad81c9c9d537a37071eaa340f666906b3c4c717147b34d4037b2961d05\",\n        \"sem\": \"3688663d5014b2a765b81fb85e5dcc16c42b54cd576e7514f09418c3f785fc50\"\n      }\n    },\n    {\n      \"nuclei\": 390,\n      \"raster_nuclei\": 390,\n      \"unrepresented_nuclei\": [],\n      \"nuclei_by_class\": {\n        \"nuclei_tumor\": 377,\n        \"nuclei_lymphocyte\": 7,\n        \"nuclei_stroma\": 1,\n        \"nuclei_endothelium\": 5\n   ","encoding":"utf8","key":"manifest.json","name":"manifest.json","size":204297,"truncated":true}],"sourceObjectKey":"puma.parquet","starterCell":"from datasets import load_dataset\nds = load_dataset(\"csmp-hub/puma\")\nprint(ds)","starterSamplePath":"examples.png","totalObjects":9},"summary":"PUMA (Panoptic segmentation of nUclei and tissue in MelAnoma) is the public training set of the PUMA challenge: 206 H&E regions of interest from melanoma, half from primary tumors…","title":"PUMA: melanoma H&E images with nucleus instances, ten nucleus classes and six tissue classes","updatedAt":"2026-11-17T23:19:56.499Z","visibility":"public","websiteUrl":"https://puma.grand-challenge.org/"}}