{"success":true,"data":{"author":"International Skin Imaging Collaboration (ISIC); repackaged by MedOtter","bucketAccessType":"public","bucketName":"MedOtter/ISIC2016","bucketPrefix":"","category":"Health & Biomedical","cloudProvider":"huggingface","createdAt":"2026-11-21T22:16:41.375Z","description":"Lesion segmentation data from the ISBI 2016 challenge Skin Lesion Analysis toward Melanoma Detection, hosted by the International Skin Imaging Collaboration (ISIC). It holds 1,279 dermoscopy photographs of pigmented skin lesions (moles, melanoma and benign nevi), 900 training and 379 test, each with one binary lesion-boundary mask traced by an expert clinician with a seeded flood-fill or manual polyline tool. Images are the original JPEGs at 0.5 to 12 megapixels and masks the original PNGs (0 background, 255 lesion); ISIC archive IDs are kept as the join key, with a flag marking images that reappear in the ISIC 2018 training set (about 90 percent do), so leakage can be filtered. Six Parquet files under data/, about 0.9 GB (measured 2026-11-21); read with Hugging Face datasets. Part 2 and Part 3 classification labels are not included. Use for skin lesion boundary segmentation benchmarks. The ISIC challenge data page lists the 2016 data under CC-0. Cite Gutman et al., arXiv:1605.01397 (2016).","displayTitle":"ISIC 2016 Part 1: dermoscopy skin lesion segmentation, 1,279 images with expert lesion masks (ISBI 2016 melanoma challenge)","huggingFaceUrl":"https://huggingface.co/datasets/MedOtter/ISIC2016","id":"55354c60-2985-5b56-926b-84b23231ff24","license":"CC0-1.0","licenseEvidenceUrl":"https://challenge.isic-archive.com/data/","objectCountEstimate":8,"probeStatus":"ok","referenceUrl":"https://arxiv.org/abs/1605.01397","region":"us","schemaCard":{"accessHints":[{"language":"python-datasets","snippet":"from datasets import load_dataset\nds = load_dataset(\"MedOtter/ISIC2016\")"},{"language":"python-hub","snippet":"from huggingface_hub import snapshot_download\nsnapshot_download(repo_id=\"MedOtter/ISIC2016\", repo_type=\"dataset\")"}],"caveats":["HuggingFace metadata: author=MedOtter, license:cc0-1.0, size_categories:1K<n<10K, format:parquet, downloads=340"],"format":"mixed","generatedAt":"2026-11-21T22:32:03.839Z","layout":{"example":"data/test-00000-of-00002.parquet","pattern":"{repoId}/{filename}"},"mountPath":"/mnt/archive","sampleHeaders":[{"bytes":"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","encoding":"hex","key":"data/test-00000-of-00002.parquet","truncatedAt":512}],"sampleObjects":["data/test-00000-of-00002.parquet","data/test-00001-of-00002.parquet","data/train-00000-of-00004.parquet","data/train-00001-of-00004.parquet","data/train-00002-of-00004.parquet"],"sidecars":[{"bytes":"---\nlicense: cc0-1.0\ntask_categories:\n  - image-segmentation\ntags:\n  - medical\n  - dermoscopy\n  - skin-lesion\n  - melanoma\n  - lesion-segmentation\n  - isic\npretty_name: ISIC 2016 (ISBI 2016) — Part 1 lesion segmentation\nsize_categories:\n  - 1K<n<10K\n---\n\n# ISIC 2016 — Part 1: Lesion Segmentation\n\n**1,279 dermoscopy images** (900 train / 379 test) of pigmented skin lesions\nwith **expert binary lesion-boundary masks**, from the ISBI 2016 challenge\n*\"Skin Lesion Analysis toward Melanoma Detection\"* hosted by the International\nSkin Imaging Collaboration (ISIC). A snapshot of the ISIC Archive.\n\n- **Modality:** dermoscopy (2D RGB JPEG), variable 0.5–12 MP resolution\n- **Organ:** skin (pigmented lesions — melanoma vs benign)\n- **Ground truth:** one binary mask per image (PNG, `0`=background,\n  `255`=lesion), traced by an expert clinician via a semi-automated\n  (seed + flood-fill) or manual (polyline) process. Single annotation tier —\n  every image has exactly one mask.\n- **Expert ceiling:** pairwise inter-observer Jaccard on 100 images of this\n  data is **≈ 0.786** (Codella et al., IBM J. Res. Dev. 2017; see\n  arXiv:1902.03368 §2.1) — treat scores near that as expert-level.\n\n## Scope\n\nThe ISBI 2016 challenge had five parts. This repository ships **Part 1\n(lesion segmentation) only.** Not included: Part 2 superpixel + dermoscopic-\nfeature classification JSONs, Part 2B dermoscopic-feature masks (globules /\nstreaks on an 807-image subset — *not* lesion segmentation ground truth), and\nPart 3/3B malignancy classification labels.\n\n## Schema\n\n| column | type | contents |\n|---|---|---|\n| `image_id` | string | ISIC Archive id, e.g. `ISIC_0000000` — the cross-challenge join key |\n| `image` | Image | original challenge JPEG, unmodified |\n| `mask` | Image | original challenge PNG, single-channel, values `{0, 255}` |\n| `in_isic2018_train` | bool | `True` iff this image is also in ISIC 2018 Task 1 **training** ground truth |\n\nSplits: `train` (900 rows), `test` (379 rows). No validation split was released\nfor this challenge.\n\n## ⚠️ Overlap with ISIC 2018 (leakage warning)\n\nThe ISIC 2016/2017/2018 challenge datasets are successive snapshots of the same\narchive and share the `ISIC_<7-digit>` id namespace. Measured against the\nofficial ISIC 2018 Task 1 training ground truth (2,594 ids):\n\n| intersection | count |\n|---|---|\n| ISIC 2016 **train** ∩ ISIC 2018 train | **806 / 900** (89.6%) |\n| ISIC 2016 **test** ∩ ISIC 2018 train | **339 / 379** (89.4%) |\n| ISIC 2016 (any) ∩ ISIC 2018 **val/test** | **0** |\n\nConsequences:\n\n- Any model **fine-tuned on ISIC 2018 training data is contaminated** for\n  evaluation on ISIC 2016 (both splits). Filter with the\n  `in_isic2018_train` column:\n  `ds.filter(lambda r: not r[\"in_isic2018_train\"])`.\n- Zero-shot evaluation on both challenges' *eval* sets never scores the same\n  image twice (2016's test set is disjoint from 2018's val/test).\n- For the 1,145 shared images the 2018 masks were slightly **revised**\n  (IoU 0.988–0.997 vs the 2016 masks) — do not mix the two years as\n  interchangeable label sources.\n\n## Provenance\n\nBuilt from the four official challenge zips (no registration required),\nfetched byte-exact against their Content-Length and matching the challenge\npaper's counts (900 train / 379 test) exactly:\n\n```\nhttps://isic-challenge-data.s3.amazonaws.com/2016/ISBI2016_ISIC_Part1_Training_Data.zip\nhttps://isic-challenge-data.s3.amazonaws.com/2016/ISBI2016_ISIC_Part1_Training_GroundTruth.zip\nhttps://isic-challenge-data.s3.amazonaws.com/2016/ISBI2016_ISIC_Part1_Test_Data.zip\nhttps://isic-challenge-data.s3.amazonaws.com/2016/ISBI2016_ISIC_Part1_Test_GroundTruth.zip\n```\n\nImages and masks are byte-identical to the originals (no re-encoding, no\nresizing). Every mask was verified single-channel with values ⊆ {0, 255},\nnon-empty, and pixel-dimension-identical to its image.\n\n## Usage\n\n```python\nfrom datasets import load_dataset\n\nds = load_dataset(\"MedOtter/ISIC2016\")          # train / test\nsample = ds[\"train\"][0]\nimage = sample[\"image\"]                          # PIL RGB\nmask = samp","encoding":"utf8","key":"README.md","name":"README.md","size":5275,"truncated":true}],"sourceObjectKey":"data/test-00000-of-00002.parquet","starterCell":"from datasets import load_dataset\nds = load_dataset(\"MedOtter/ISIC2016\")\nprint(ds)","starterSamplePath":"data/test-00000-of-00002.parquet","totalObjects":8},"summary":"Lesion segmentation data from the ISBI 2016 challenge Skin Lesion Analysis toward Melanoma Detection, hosted by the International Skin Imaging Collaboration (ISIC).","title":"ISIC 2016 Part 1: dermoscopy skin lesion segmentation, 1,279 images with expert lesion masks (ISBI 2016 melanoma challenge)","updatedAt":"2026-11-22T01:03:22.221Z","visibility":"public","websiteUrl":"https://challenge.isic-archive.com/data/"}}