{"success":true,"data":{"author":"Hakai Institute; University of California Santa Cruz","bucketAccessType":"public","bucketName":"HakaiInstitute/mussel-seg-1024-1024","bucketPrefix":"","category":"Environmental Science","cloudProvider":"huggingface","createdAt":"2026-11-21T20:45:58.691Z","description":"A semantic segmentation dataset of rocky intertidal mussel habitat from very high resolution drone (UAV) imagery collected by the Hakai Institute and the University of California Santa Cruz between 2021 and 2023 on the central coast of British Columbia, Canada, and in California, USA, covering about 3,500 square meters. Each of 9,972 GeoTIFF tiles (1024 by 1024 pixels, 50 percent overlap, mean ground sampling distance 0.45 cm) is paired with a pixel mask that labels mussel bed (about 13 percent of pixels) or background, made by manual heads-up digitizing with manual verification. Data are WebDataset tar shards split into train, validation and test folders (42 GB), loaded with the datasets library (load_dataset with split train, validation or test; keys img.tif and seg.tif) or streamed. Intended for training models that map mussel beds for coastal ecology and intertidal monitoring. Released by the Hakai Institute under CC BY 4.0; cite the dataset, DOI 10.57967/hf/2760.","displayTitle":"MusselSeg: Drone Imagery Semantic Segmentation of Rocky Intertidal Mussel Beds, British Columbia and California (Hakai Institute)","doi":"10.57967/hf/2760","huggingFaceUrl":"https://huggingface.co/datasets/HakaiInstitute/mussel-seg-1024-1024","id":"b734251f-0011-5d87-bd3e-18770e8e69cb","license":"CC-BY-4.0","licenseEvidenceUrl":"https://huggingface.co/datasets/HakaiInstitute/mussel-seg-1024-1024","objectCountEstimate":17,"probeStatus":"ok","referenceUrl":"https://huggingface.co/datasets/HakaiInstitute/mussel-seg-1024-1024","region":"us","schemaCard":{"accessHints":[{"language":"python-datasets","snippet":"from datasets import load_dataset\nds = load_dataset(\"HakaiInstitute/mussel-seg-1024-1024\")"},{"language":"python-hub","snippet":"from huggingface_hub import snapshot_download\nsnapshot_download(repo_id=\"HakaiInstitute/mussel-seg-1024-1024\", repo_type=\"dataset\")"}],"caveats":["HuggingFace metadata: author=HakaiInstitute, license:cc-by-4.0, size_categories:1K<n<10K, format:webdataset, downloads=271"],"format":"mixed","generatedAt":"2026-11-21T20:48:47.198Z","layout":{"example":"test/000000.tar","pattern":"{repoId}/{filename}"},"mountPath":"/mnt/archive","sampleHeaders":[{"bytes":"2e2f2e2f40506178486561646572000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000030303030303030003030303030303000303030303030300030303030303030303033340030303030303030303030300030313032313200207800000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000757374617200303000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000","encoding":"hex","key":"test/000000.tar","truncatedAt":512}],"sampleObjects":["test/000000.tar","test/000001.tar","test/000002.tar","test/000003.tar","test/000004.tar"],"sidecars":[{"bytes":"---\nlicense: cc-by-4.0\ntask_categories:\n- image-segmentation\nlanguage:\n- en\ntags:\n- image\n- geospatial\n- biology\n- aerial imagery\n- remote sensing\npretty_name: MusselSeg\nsize_categories:\n- 1K<n<10K\n---\n\n# MusselSeg: Semantic Segmentation for Rocky Intertidal Mussel Habitat\n\n## Dataset description\n\nMusselSeg is a large-scale dataset for semantic segmentation of mussel habitat using high resolution drone imagery. It covers coastal mussel habitat located on the central coast of British Columbia, Canada, as well as areas in California, USA and provides pixel-wise annotation for mussel beds.\n\n- **Source:** Imagery collected by the Hakai Institute and University of California Santa Cruz\n- **Geographic Coverage:** ~3500m<sup>2</sup>\n- **Time Period:** 2021-2023\n\n## Task description\n\nThe dataset is designed for semantic segmentation of mussel habitat in aerial imagery. The task involves assigning each pixel in the image to either the \"mussel\" class or \"background\" (i.e. not mussels) class.\n\n## Usage\n\n### Download and iterate\n\nInstall the HuggingFace datasets library ([instructions](https://huggingface.co/docs/datasets/en/installation))\n\n```python\nfrom datasets import load_dataset\n\ntrain_dataset = load_dataset(\"HakaiInstitute/mussel-seg-1024-1024\", split=\"train\")\nval_dataset = load_dataset(\"HakaiInstitute/mussel-seg-1024-1024\", split=\"validation\")\ntest_dataset  = load_dataset(\"HakaiInstitute/mussel-seg-1024-1024\", split=\"test\")\n\nfor sample in train_dataset:\n    x = sample[\"img.tif\"]\n    y = sample[\"seg.tif\"]\n    # x and y are `PIL.Image` instances, ready to feed into a training loop, PyTorch dataloader, etc.\n\n    # ...\n```\n\n### Streaming from HuggingFace\n\nThis data is released as a WebDatasets, which makes it possible to use the data without downloading it in advance. \nFor instructions on how to do this, please see [WebDataset](https://huggingface.co/docs/hub/en/datasets-webdataset)\n\n## Data characteristics\n\n- **Image Format:** GeoTiff\n- **Resolution:** mean=0.45cm, stdev=0.20cm\n- **Tile Size:** 1024x1024 pixels with 50% overlap\n- **Number of Tiles:** 9972 image and label pairs\n- **Total Dataset Size:** 42G\n\n## Annotation details\n- **Method:** Manual heads-up digitizing with manual verification\n- **Format:** Pixel-wise labels stored as separate mask images\n- **Labelling Convention:** Each pixel assigned a single class label\n\n## Class distribution\n\n| Class ID | Class Name | Description        | Percentage |\n| :------- | :--------- | :----------------- | :--------: |\n| 0        | Background | Unclassified areas | 87%        |\n| 1        | Mussels    | Mussel bed         | 13%        |\n\n## Split information\n\n| Split      | Data Percentage | Tiles Count | Mussel Pixels |\n| :--------- | --------------: | ----------: | ------------: |\n| Train      |             48% |        4834 |         18.4% |\n| Validation |             13% |        1277 |         16.7% |\n| Test       |             39% |        3861 |          4.3% |\n\nTrain and Validation split tiles all contain at least 1 pixel in each class. For the Test split, some tiles are entirely the background class. If ignoring the test split tiles which contain only background pixels, the split percentages instead become 70/17/13 for the train/validation/test splits, respectively.\n\nSplits are created such that tiles from the same source orthomosaic image are not divided across different splits. That is, all tiles from the same drone flight are present only in a single split.\n\n## Preprocessing\n\n1. Orthorectification applied to raw imagery\n2. Tiles extracted with 50% overlap\n3. Tiles with no mussels present eliminated for the Train and Validation splits\n\n## Licensing information\n\nThis dataset is released under the Creative Commons Attribution 4.0 License (CC BY 4.0).\n\n## Ethical considerations\n\n- No identifiable individuals are present in imagery\n- Minimized impact on wildlife and sensitive habitats\n- Engaged with local First Nations in planning aerial surveys\n\n## Citation information\n\nIf you use this dataset in your research,  please cite:\n\n```\n@misc{denouden2024musselseg,\n  author = {Denou","encoding":"utf8","key":"README.md","name":"README.md","size":4799,"truncated":true}],"sourceObjectKey":"test/000000.tar","starterCell":"from huggingface_hub import hf_hub_download\npath = hf_hub_download(repo_id=\"HakaiInstitute/mussel-seg-1024-1024\", filename=\"test/000000.tar\", repo_type=\"dataset\")\nprint(path)","starterSamplePath":"test/000000.tar","totalObjects":17},"summary":"A semantic segmentation dataset of rocky intertidal mussel habitat from very high resolution drone (UAV) imagery collected by the Hakai Institute and the University of California…","title":"MusselSeg: Drone Imagery Semantic Segmentation of Rocky Intertidal Mussel Beds, British Columbia and California (Hakai Institute)","updatedAt":"2026-11-21T21:54:11.092Z","visibility":"public","websiteUrl":"https://hakai.org"}}