README.md

---
pretty_name: Reef Support Marine Data
license: other
license_name: mixed-open
license_link: LICENSE
language:
- en
task_categories:
- image-segmentation
- object-detection
- image-classification
task_ids:
- semantic-segmentation
- instance-segmentation
- multi-class-image-classification
annotations_creators:
- expert-generated
- found
source_datasets:
- original
- extended|NMFS-OSI/NOAA-PIFSC-ESD-CORAL-Bleaching-Dataset
- extended|LiamLian0727/UIIS
- extended|LiamLian0727/UIIS10K
- extended|LiamLian0727/USIS10K
size_categories:
- 10K<n<100K
tags:
- marine
- coral-reef
- underwater
- biology
- ecology
- conservation
- coral
- benthic
- fish
- bleaching
- semantic-segmentation
- instance-segmentation
- object-detection
- image-classification
- taxonomy
- reef-support
configs:
- config_name: coral-masks
  data_files:
  - split: train
    path: data/coral-masks/train-*.parquet
  - split: validation
    path: data/coral-masks/validation-*.parquet
  - split: test
    path: data/coral-masks/test-*.parquet
- config_name: scene-masks
  data_files:
  - split: train
    path: data/scene-masks/train-*.parquet
  - split: validation
    path: data/scene-masks/validation-*.parquet
  - split: test
    path: data/scene-masks/test-*.parquet
- config_name: instance-masks
  data_files:
  - split: train
    path: data/instance-masks/train-*.parquet
  - split: validation
    path: data/instance-masks/validation-*.parquet
  - split: test
    path: data/instance-masks/test-*.parquet
- config_name: fish-boxes
  data_files:
  - split: train
    path: data/fish-boxes/train-*.parquet
  - split: validation
    path: data/fish-boxes/validation-*.parquet
  - split: test
    path: data/fish-boxes/test-*.parquet
- config_name: images
  data_files:
  - split: train
    path: data/images/train-*.parquet
  - split: validation
    path: data/images/validation-*.parquet
  - split: test
    path: data/images/test-*.parquet
- config_name: metadata
  default: true
  data_files:
  - split: train
    path: data/metadata/train-*.parquet
  - split: validation
    path: data/metadata/validation-*.parquet
  - split: test
    path: data/metadata/test-*.parquet
- config_name: points
  data_files:
  - split: train
    path: data/points/train-*.parquet
  - split: validation
    path: data/points/validation-*.parquet
  - split: test
    path: data/points/test-*.parquet
- config_name: bleaching
  data_files:
  - split: train
    path: data/bleaching/train-*.parquet
  - split: validation
    path: data/bleaching/validation-*.parquet
  - split: test
    path: data/bleaching/test-*.parquet
- config_name: boxes
  data_files:
  - split: train
    path: data/boxes/train-*.parquet
  - split: validation
    path: data/boxes/validation-*.parquet
  - split: test
    path: data/boxes/test-*.parquet
- config_name: semseg
  data_files:
  - split: train
    path: data/semseg/train-*.parquet
  - split: validation
    path: data/semseg/validation-*.parquet
  - split: test
    path: data/semseg/test-*.parquet
- config_name: condition-masks
  data_files:
  - split: train
    path: data/condition-masks/train-*.parquet
- config_name: instances
  data_files:
  - split: train
    path: data/instances/train-*.parquet
  - split: validation
    path: data/instances/validation-*.parquet
  - split: test
    path: data/instances/test-*.parquet
- config_name: benthic-cover
  data_files:
  - split: train
    path: data/benthic-cover/train-*.parquet
- config_name: taxonomy
  data_files:
  - split: train
    path: data/taxonomy/train-*.parquet
- config_name: label-map
  data_files:
  - split: train
    path: data/label-map/train-*.parquet
- config_name: sources
  data_files:
  - split: train
    path: data/sources/train-*.parquet
dataset_info:
- config_name: images
  features:
  - name: image_sha256
    dtype: string
  - name: image
    dtype: image
  - name: bytes_sha256
    dtype: string
  - name: source_id
    dtype: string
  - name: source_ids
    dtype: string
  - name: split_group
    dtype: string
  - name: split
    dtype: string
  splits:
  - name: train
    num_bytes: 12657378303
    num_examples: 22193
  - name: validation
    num_bytes: 2500373334
    num_examples: 4562
  - name: test
    num_bytes: 3822054346
    num_examples: 5445
  download_size: 18979805983
  dataset_size: 18979805983
- config_name: metadata
  features:
  - name: image_sha256
    dtype: string
  - name: source_id
    dtype: string
  - name: source_version
    dtype: string
  - name: license
    dtype: string
  - name: licence_class
    dtype: string
  - name: attribution
    dtype: string
  - name: split_group
    dtype: string
  - name: split
    dtype: string
  - name: upstream_id
    dtype: string
  - name: upstream_url
    dtype: string
  - name: fetch_date
    dtype: string
  - name: lineage_root_digest
    dtype: string
  - name: capture_datetime
    dtype: string
  - name: lat
    dtype: float64
  - name: lon
    dtype: float64
  - name: geo_precision
    dtype: string
  - name: geo_source
    dtype: string
  - name: gps_precision_m
    dtype: string
  - name: depth_m
    dtype: string
  - name: depth_source
    dtype: string
  - name: platform
    dtype: string
  - name: camera
    dtype: string
  - name: meow_realm
    dtype: string
  - name: meow_province
    dtype: string
  - name: meow_ecoregion
    dtype: string
  - name: depth_zone
    dtype: string
  - name: habitat
    dtype: string
  - name: location_generalized
    dtype: bool
  - name: min_side
    dtype: string
  - name: q_blur
    dtype: string
  - name: q_clip_lo
    dtype: string
  - name: q_clip_hi
    dtype: string
  - name: q_uiqm
    dtype: string
  - name: q_entropy
    dtype: string
  - name: q_blank
    dtype: string
  - name: quality_flags
    dtype: string
  - name: privacy_flag
    dtype: string
  - name: face_score
    dtype: string
  - name: date
    dtype: string
  - name: date_precision
    dtype: string
  - name: year
    dtype: string
  - name: site_id
    dtype: string
  - name: quadrat_id
    dtype: string
  - name: survey_id
    dtype: string
  - name: region
    dtype: string
  - name: coord_precision_m
    dtype: string
  - name: date_source
    dtype: string
  - name: license_url
    dtype: string
  - name: rights_holder
    dtype: string
  - name: license_verified_by
    dtype: string
  - name: license_verified_on
    dtype: string
  - name: provenance_url
    dtype: string
  - name: bytes_config
    dtype: string
  - name: bytes_sha256
    dtype: string
  - name: taxonomy_version
    dtype: string
  splits:
  - name: train
    num_bytes: 5316069
    num_examples: 40183
  - name: validation
    num_bytes: 1377058
    num_examples: 8096
  - name: test
    num_bytes: 2200460
    num_examples: 17077
  download_size: 8893587
  dataset_size: 8893587
- config_name: points
  features:
  - name: image_sha256
    dtype: string
  - name: source_id
    dtype: string
  - name: source_version
    dtype: string
  - name: ann_id
    dtype: string
  - name: label_native
    dtype: string
  - name: label_native_id
    dtype: string
  - name: label_set
    dtype: string
  - name: taxon_node_id
    dtype: string
  - name: form_node_id
    dtype: string
  - name: condition_node_id
    dtype: string
  - name: taxon_rank
    dtype: string
  - name: worms_aphia_id
    dtype: int64
  - name: rs_benthic_code
    dtype: string
  - name: match_type
    dtype: string
  - name: annotator_type
    dtype: string
  - name: annotator_detail
    dtype: string
  - name: ann_license
    dtype: string
  - name: ann_attribution
    dtype: string
  - name: confidence
    dtype: string
  - name: upstream_split
    dtype: string
  - name: label_status
    dtype: string
  - name: attrs
    dtype: string
  - name: x
    dtype: float64
  - name: y
    dtype: float64
  - name: x_px
    dtype: int64
  - name: y_px
    dtype: int64
  - name: split
    dtype: string
  - name: taxonomy_version
    dtype: string
  - name: fidelity
    dtype: string
  - name: bytes_config
    dtype: string
  - name: quadrat_id
    dtype: string
  - name: label_native_group
    dtype: string
  - name: x_norm
    dtype: float64
  - name: y_norm
    dtype: float64
  splits:
  - name: train
    num_bytes: 14459389
    num_examples: 559490
  - name: validation
    num_bytes: 3404392
    num_examples: 108730
  - name: test
    num_bytes: 5646065
    num_examples: 190100
  download_size: 23509846
  dataset_size: 23509846
- config_name: bleaching
  features:
  - name: image_sha256
    dtype: string
  - name: source_id
    dtype: string
  - name: source_version
    dtype: string
  - name: ann_id
    dtype: string
  - name: label_native
    dtype: string
  - name: label_native_id
    dtype: string
  - name: label_set
    dtype: string
  - name: taxon_node_id
    dtype: string
  - name: form_node_id
    dtype: string
  - name: condition_node_id
    dtype: string
  - name: taxon_rank
    dtype: string
  - name: worms_aphia_id
    dtype: string
  - name: rs_benthic_code
    dtype: string
  - name: match_type
    dtype: string
  - name: annotator_type
    dtype: string
  - name: annotator_detail
    dtype: string
  - name: ann_license
    dtype: string
  - name: ann_attribution
    dtype: string
  - name: confidence
    dtype: string
  - name: upstream_split
    dtype: string
  - name: label_status
    dtype: string
  - name: attrs
    dtype: string
  - name: label_key
    dtype: string
  - name: near_dup_group
    dtype: string
  - name: near_dup_images
    dtype: int64
  - name: split
    dtype: string
  - name: taxonomy_version
    dtype: string
  - name: fidelity
    dtype: string
  - name: bytes_config
    dtype: string
  - name: sha256
    dtype: string
  - name: label_origin
    dtype: string
  - name: licence_class
    dtype: string
  - name: native_label
    dtype: string
  - name: canonical_condition
    dtype: string
  - name: bleaching_condition
    dtype: string
  - name: coral_health
    dtype: string
  - name: evidence
    dtype: string
  - name: pixel_count
    dtype: int64
  splits:
  - name: train
    num_bytes: 2417333
    num_examples: 19631
  - name: validation
    num_bytes: 556933
    num_examples: 3823
  - name: test
    num_bytes: 560886
    num_examples: 3889
  download_size: 3535152
  dataset_size: 3535152
- config_name: boxes
  features:
  - name: image_sha256
    dtype: string
  - name: source_id
    dtype: string
  - name: source_version
    dtype: string
  - name: ann_id
    dtype: string
  - name: label_native
    dtype: string
  - name: label_native_id
    dtype: string
  - name: label_set
    dtype: string
  - name: taxon_node_id
    dtype: string
  - name: form_node_id
    dtype: string
  - name: condition_node_id
    dtype: string
  - name: taxon_rank
    dtype: string
  - name: worms_aphia_id
    dtype: int64
  - name: rs_benthic_code
    dtype: string
  - name: match_type
    dtype: string
  - name: annotator_type
    dtype: string
  - name: annotator_detail
    dtype: string
  - name: ann_license
    dtype: string
  - name: ann_attribution
    dtype: string
  - name: confidence
    dtype: string
  - name: upstream_split
    dtype: string
  - name: label_status
    dtype: string
  - name: attrs
    dtype: string
  - name: x_min
    dtype: float64
  - name: y_min
    dtype: float64
  - name: x_max
    dtype: float64
  - name: y_max
    dtype: float64
  - name: x_min_px
    dtype: int64
  - name: y_min_px
    dtype: int64
  - name: x_max_px
    dtype: int64
  - name: y_max_px
    dtype: int64
  - name: is_crowd
    dtype: bool
  - name: img_w
    dtype: int64
  - name: img_h
    dtype: int64
  - name: modality
    dtype: string
  - name: split
    dtype: string
  - name: taxonomy_version
    dtype: string
  - name: fidelity
    dtype: string
  - name: bytes_config
    dtype: string
  - name: licence_class
    dtype: string
  splits:
  - name: train
    num_bytes: 2149451
    num_examples: 49038
  - name: validation
    num_bytes: 522180
    num_examples: 8417
  - name: test
    num_bytes: 1645129
    num_examples: 36503
  download_size: 4316760
  dataset_size: 4316760
- config_name: semseg
  features:
  - name: image_sha256
    dtype: string
  - name: source_id
    dtype: string
  - name: source_version
    dtype: string
  - name: ann_id
    dtype: string
  - name: label_native
    dtype: string
  - name: label_native_id
    dtype: string
  - name: label_set
    dtype: string
  - name: taxon_node_id
    dtype: string
  - name: form_node_id
    dtype: string
  - name: condition_node_id
    dtype: string
  - name: taxon_rank
    dtype: string
  - name: worms_aphia_id
    dtype: string
  - name: rs_benthic_code
    dtype: string
  - name: match_type
    dtype: string
  - name: annotator_type
    dtype: string
  - name: annotator_detail
    dtype: string
  - name: ann_license
    dtype: string
  - name: ann_attribution
    dtype: string
  - name: confidence
    dtype: string
  - name: upstream_split
    dtype: string
  - name: label_status
    dtype: string
  - name: attrs
    dtype: string
  - name: mask_kind
    dtype: string
  - name: mask_ref
    dtype: string
  - name: rle
    dtype: string
  - name: polygon
    dtype: string
  - name: class_map
    dtype: string
  - name: ignore_value
    dtype: string
  - name: class_counts
    dtype: string
  - name: canonical_class_counts
    dtype: string
  - name: instance_id
    dtype: string
  - name: mask_config
    dtype: string
  - name: palette
    dtype: string
  - name: split
    dtype: string
  - name: taxonomy_version
    dtype: string
  - name: fidelity
    dtype: string
  - name: bytes_config
    dtype: string
  - name: sha256
    dtype: string
  - name: label_origin
    dtype: string
  - name: mask_key
    dtype: string
  - name: hf_class_map
    dtype: string
  - name: mask_encoding
    dtype: string
  - name: mask_parts
    dtype: string
  - name: mask
    dtype: image
  - name: mask_bytes_sha256
    dtype: string
  splits:
  - name: train
    num_bytes: 946608
    num_examples: 2445
  - name: validation
    num_bytes: 138107
    num_examples: 295
  - name: test
    num_bytes: 238723
    num_examples: 559
  download_size: 1323438
  dataset_size: 1323438
- config_name: condition-masks
  features:
  - name: image_sha256
    dtype: string
  - name: source_id
    dtype: string
  - name: split
    dtype: string
  - name: class_map
    dtype: string
  - name: ignore_value
    dtype: string
  - name: attrs
    dtype: string
  - name: mask
    dtype: image
  - name: mask_bytes_sha256
    dtype: string
  - name: mask_config
    dtype: string
- config_name: instances
  features:
  - name: image_sha256
    dtype: string
  - name: source_id
    dtype: string
  - name: source_version
    dtype: string
  - name: ann_id
    dtype: string
  - name: label_native
    dtype: string
  - name: label_native_id
    dtype: string
  - name: label_set
    dtype: string
  - name: taxon_node_id
    dtype: string
  - name: form_node_id
    dtype: string
  - name: condition_node_id
    dtype: string
  - name: taxon_rank
    dtype: string
  - name: worms_aphia_id
    dtype: int64
  - name: rs_benthic_code
    dtype: string
  - name: match_type
    dtype: string
  - name: annotator_type
    dtype: string
  - name: annotator_detail
    dtype: string
  - name: ann_license
    dtype: string
  - name: ann_attribution
    dtype: string
  - name: confidence
    dtype: string
  - name: upstream_split
    dtype: string
  - name: label_status
    dtype: string
  - name: attrs
    dtype: string
  - name: mask_kind
    dtype: string
  - name: mask_ref
    dtype: string
  - name: rle
    dtype: string
  - name: polygon
    dtype: string
  - name: class_map
    dtype: string
  - name: ignore_value
    dtype: string
  - name: class_counts
    dtype: string
  - name: canonical_class_counts
    dtype: string
  - name: instance_id
    dtype: int64
  - name: modality
    dtype: string
  - name: mask_config
    dtype: string
  - name: split
    dtype: string
  - name: taxonomy_version
    dtype: string
  - name: fidelity
    dtype: string
  - name: bytes_config
    dtype: string
  - name: sha256
    dtype: string
  - name: licence_class
    dtype: string
  - name: coarse
    dtype: string
  splits:
  - name: train
    num_bytes: 51356372
    num_examples: 45869
  - name: validation
    num_bytes: 10390132
    num_examples: 7664
  - name: test
    num_bytes: 42615939
    num_examples: 35612
  download_size: 104362443
  dataset_size: 104362443
- config_name: benthic-cover
  features:
  - name: sha256
    dtype: string
  - name: source_id
    dtype: string
  - name: label_origin
    dtype: string
  - name: benthic_cover
    dtype: string
  - name: benthic_dominant
    dtype: string
  - name: benthic_present
    dtype: string
  - name: label_status
    dtype: string
  splits:
  - name: train
    num_bytes: 287794
    num_examples: 3298
  download_size: 287794
  dataset_size: 287794
- config_name: taxonomy
  features:
  - name: kind
    dtype: string
  - name: key
    dtype: string
  - name: name
    dtype: string
  - name: parent
    dtype: string
  - name: rank
    dtype: string
  - name: axis
    dtype: string
  - name: aphia_id
    dtype: int64
  - name: non_taxon
    dtype: bool
  - name: fidelity
    dtype: string
  - name: targets
    dtype: string
  - name: taxonomy_version
    dtype: string
  splits:
  - name: train
    num_bytes: 87382
    num_examples: 2937
  download_size: 87382
  dataset_size: 87382
- config_name: label-map
  features:
  - name: source_id
    dtype: string
  - name: crosswalk_id
    dtype: string
  - name: native_label
    dtype: string
  - name: native_label_id
    dtype: string
  - name: pixel_value
    dtype: int64
  - name: taxon_node_id
    dtype: string
  - name: form_node_id
    dtype: string
  - name: condition_node_id
    dtype: string
  - name: match_type
    dtype: string
  - name: fidelity
    dtype: string
  - name: mapped
    dtype: bool
  - name: note
    dtype: string
  - name: taxonomy_version
    dtype: string
  splits:
  - name: train
    num_bytes: 15366
    num_examples: 306
  download_size: 15366
  dataset_size: 15366
- config_name: sources
  features:
  - name: source_id
    dtype: string
  - name: version
    dtype: string
  - name: root_digest
    dtype: string
  - name: licence
    dtype: string
  - name: licence_url
    dtype: string
  - name: ann_licence
    dtype: string
  - name: attribution
    dtype: string
  - name: rights_holder
    dtype: string
  - name: verified_by
    dtype: string
  - name: verified_on
    dtype: string
  - name: provenance_url
    dtype: string
  - name: licence_class
    dtype: string
  - name: n_images
    dtype: int64
  - name: configs
    dtype: string
  - name: n_labels
    dtype: int64
  - name: mapped_pct
    dtype: float64
  - name: known_defects
    dtype: string
  - name: taxonomy_version
    dtype: string
  splits:
  - name: train
    num_bytes: 10152
    num_examples: 14
  download_size: 10152
  dataset_size: 10152
---

# Reef Support Marine Data

Underwater imagery with masks, boxes, points and image labels for coral-reef computer vision, harmonised into one label space, with the licence and attribution of every row recorded.

> **Open flavour**: every source here allows commercial use (per-source licences and attribution below).

## Dataset summary

Reef Support Marine Data v1.1.0 adds 11 configurations to the frozen v1.0 release, additively: image bytes once per `image_sha256`, per-image metadata, point, box, mask and bleaching labels, benthic cover and sites, and the tables that describe the label space (`taxonomy`, `label-map`) and the sources. Every row carries its `image_sha256`, so any label reaches its pixels in one join, and `licence` and attribution stay attached to the pixels.

| Field | Value |
|---|---|
| Configurations | 12: `images`, `metadata`, `points`, `bleaching`, `boxes`, `semseg`, `condition-masks`, `instances`, `benthic-cover`, `taxonomy`, `label-map`, `sources` |
| Images (`metadata` rows) | 65,356 |
| Sources | 14 |
| Format | Parquet shards, streamable; images and new masks embedded once |
| Size of the new configs | 19.1 GB |
| Version | v1.1.1 (v1.1.0 plus `condition-masks`), built 2026-10-17 |
| Code | [github.com/reefsupport/marine-data](https://github.com/reefsupport/marine-data) |

## Supported tasks

| Config | Task | Rows | Sources | Notes |
|---|---|---|---|---|
| `images` | Pixels for retrieval, pretraining and image classification | 32,200 | `noaa-pifsc-bleaching`, `reef-support-seaview-labels`, `roboflow-aquarium`, `roboflow-coral-bleaching-final-v6i`, `roboflow-coral-bleaching-general-v1-yolov8s`, `roboflow-coral-classification-copy-changed-v13i`, `roboflow-coral-reef-classification-v3i`, `seaview-survey-imagery` | Bytes once per `image_sha256`; images a v1.0 config already embeds are not repeated (`metadata.bytes_config` names the config that holds each image). |
| `metadata` | Survey and site analytics, joins, retrieval | 65,356 | `coralscapes`, `noaa-pifsc-bleaching`, `reef-support-benthic-own`, `reef-support-seaview-labels`, `roboflow-aquarium`, `roboflow-coral-bleaching-final-v6i`, `roboflow-coral-bleaching-general-v1-yolov8s`, `roboflow-coral-classification-copy-changed-v13i`, `roboflow-coral-reef-classification-v3i`, `seaview-survey-imagery`, `suim`, `uiis`, `uiis10k`, `usis10k` | One row per image, v1.0 images included; no pixels. Default config. |
| `points` | Point-based benthic cover | 858,320 | `seaview-survey-imagery` | One row per labelled point; label-only, no image. |
| `bleaching` | Image classification (bleaching condition) | 27,343 | `noaa-pifsc-bleaching`, `reef-support-bleaching`, `roboflow-coral-bleaching-final-v6i`, `roboflow-coral-bleaching-general-v1-yolov8s`, `roboflow-coral-classification-copy-changed-v13i`, `roboflow-coral-reef-classification-v3i` | One row per (image, label). |
| `boxes` | Object detection | 93,958 | `roboflow-aquarium`, `uiis`, `uiis10k`, `usis10k` | One row per box; label-only. |
| `semseg` | Semantic segmentation | 3,299 | `coralscapes`, `reef-support-benthic-own` | One row per (image, source) mask. New masks are embedded as palette PNG; v1.0 masks are referenced by `mask_ref` / `mask_config`. |
| `instances` | Instance segmentation | 89,145 | `uiis`, `uiis10k`, `usis10k` | One row per instance. No mask pixels: each row holds the instance geometry (`polygon`, `rle` or a `mask_ref` pointer), the label and its taxonomy nodes; the masks stay in the v1.0 `instance-masks` config. |
| `benthic-cover` | Benthic cover | 3,298 | `coralscapes`, `reef-support-benthic-own` | Derived per image or survey from points and masks. |
| `taxonomy` | Label space | 2,937 | – | Taxonomy nodes and the edges of the crosswalks in this repo. |
| `label-map` | Label harmonisation | 306 | `coralscapes`, `noaa-pifsc-bleaching`, `reef-support-benthic-own`, `reef-support-seaview-labels`, `roboflow-aquarium`, `roboflow-coral-bleaching-final-v6i`, `roboflow-coral-bleaching-general-v1-yolov8s`, `roboflow-coral-classification-copy-changed-v13i`, `roboflow-coral-reef-classification-v3i`, `seaview-survey-imagery`, `suim`, `uiis`, `uiis10k`, `usis10k` | One row per (source, native label or pixel value) with its taxonomy nodes and match type. |
| `sources` | Provenance, licence and attribution | 14 | `coralscapes`, `noaa-pifsc-bleaching`, `reef-support-benthic-own`, `reef-support-seaview-labels`, `roboflow-aquarium`, `roboflow-coral-bleaching-final-v6i`, `roboflow-coral-bleaching-general-v1-yolov8s`, `roboflow-coral-classification-copy-changed-v13i`, `roboflow-coral-reef-classification-v3i`, `seaview-survey-imagery`, `suim`, `uiis`, `uiis10k`, `usis10k` | One row per source: licence, attribution, rights holder, verification, counts. |

## Quick start

```python
from datasets import load_dataset

meta = load_dataset("reefsupport/marine-data", "metadata", split="train")  # default, no pixels
images = load_dataset("reefsupport/marine-data", "images", split="train", streaming=True)
```

Every label table joins to `metadata` and `images` on `image_sha256`; `metadata.bytes_config` names the config that holds the pixels.

```python
import duckdb

duckdb.sql("""
    SELECT p.image_sha256, p.label_native, m.bytes_config
    FROM 'hf://datasets/reefsupport/marine-data/data/points/*.parquet' p
    JOIN 'hf://datasets/reefsupport/marine-data/data/metadata/*.parquet' m USING (image_sha256)
    LIMIT 5
""").show()

duckdb.sql("""
    SELECT * FROM 'hf://datasets/reefsupport/marine-data/data/label-map/*.parquet'
    WHERE match_type = 'unmapped' LIMIT 5
""").show()
```

## Dataset structure

- `images` (7 columns): `image_sha256`, `image`, `bytes_sha256`, `source_id`, `source_ids`, `split_group`, `split`.
- `metadata` (55 columns): `image_sha256`, `source_id`, `source_version`, `license`, `licence_class`, `attribution`, `split_group`, `split`, `upstream_id`, `upstream_url`, `fetch_date`, `lineage_root_digest`, `capture_datetime`, `lat`, and 41 more.
- `points` (34 columns): `image_sha256`, `source_id`, `source_version`, `ann_id`, `label_native`, `label_native_id`, `label_set`, `taxon_node_id`, `form_node_id`, `condition_node_id`, `taxon_rank`, `worms_aphia_id`, `rs_benthic_code`, `match_type`, and 20 more.
- `bleaching` (38 columns): `image_sha256`, `source_id`, `source_version`, `ann_id`, `label_native`, `label_native_id`, `label_set`, `taxon_node_id`, `form_node_id`, `condition_node_id`, `taxon_rank`, `worms_aphia_id`, `rs_benthic_code`, `match_type`, and 24 more.
- `boxes` (39 columns): `image_sha256`, `source_id`, `source_version`, `ann_id`, `label_native`, `label_native_id`, `label_set`, `taxon_node_id`, `form_node_id`, `condition_node_id`, `taxon_rank`, `worms_aphia_id`, `rs_benthic_code`, `match_type`, and 25 more.
- `semseg` (45 columns): `image_sha256`, `source_id`, `source_version`, `ann_id`, `label_native`, `label_native_id`, `label_set`, `taxon_node_id`, `form_node_id`, `condition_node_id`, `taxon_rank`, `worms_aphia_id`, `rs_benthic_code`, `match_type`, and 31 more.
- `condition-masks` (9 columns): `image_sha256`, `source_id`, `split`, `class_map`, `ignore_value`, `attrs`, `mask`, `mask_bytes_sha256`, `mask_config`.
- `instances` (40 columns): `image_sha256`, `source_id`, `source_version`, `ann_id`, `label_native`, `label_native_id`, `label_set`, `taxon_node_id`, `form_node_id`, `condition_node_id`, `taxon_rank`, `worms_aphia_id`, `rs_benthic_code`, `match_type`, and 26 more.
- `benthic-cover` (7 columns): `sha256`, `source_id`, `label_origin`, `benthic_cover`, `benthic_dominant`, `benthic_present`, `label_status`.
- `taxonomy` (11 columns): `kind`, `key`, `name`, `parent`, `rank`, `axis`, `aphia_id`, `non_taxon`, `fidelity`, `targets`, `taxonomy_version`.
- `label-map` (13 columns): `source_id`, `crosswalk_id`, `native_label`, `native_label_id`, `pixel_value`, `taxon_node_id`, `form_node_id`, `condition_node_id`, `match_type`, `fidelity`, `mapped`, `note`, `taxonomy_version`.
- `sources` (18 columns): `source_id`, `version`, `root_digest`, `licence`, `licence_url`, `ann_licence`, `attribution`, `rights_holder`, `verified_by`, `verified_on`, `provenance_url`, `licence_class`, `n_images`, `configs`, and 4 more.

Keys: the image key is `image_sha256` (64 lowercase hex, the staged checksum, never recomputed from re-encoded bytes); `metadata.bytes_sha256` is the hash of the embedded bytes. The annotation key is `image_sha256` + `source_id` + `source_version` + `ann_id`. `metadata.bytes_config` names the config that holds each image's pixels: a v1.0 config or `images`. The Hub splits are `train`, `validation`, `test` and `test_ood`.

**Masks.** A `semseg` row names a v1.0 mask config in `mask_config` when that frozen config holds the image's mask, and the mask stays there; otherwise the mask is embedded in the row as a palette PNG. The `instances` table carries no masks: each row holds the instance's `polygon` or `rle` (or a `mask_ref` pointer), `instance_id`, the native label with its taxonomy nodes and `match_type`, `annotator_type` and `ann_license`, and the instance masks stay in the v1.0 `instance-masks` config.

**Condition masks (v1.1.1).** `condition-masks` holds the Tayrona bleaching rasters of `reef-support-bleaching` as palette PNGs, `train` only, one row per image (hidden-test images withheld). The image bytes are not repeated: `metadata.bytes_config` names the config that holds them. Pixel `255` is ignore (unlabelled in the source raster, value 0), `1` is `bleached` (condition `BLEACHED`) and `2` is `non_bleached` (condition `HEALTHY`: the source only recorded bleached or not). `class_map` keeps the source pixel values; `attrs.class_resolution` carries the condition node. The v1.1.0 `label-map` has no rows for this source; this note is the definition.

## Label space

Labels are harmonised into the pinned taxonomy release (`taxonomy_version` is stamped on every row of every new table). Three axes per row: `taxon_node_id`, `form_node_id` and `condition_node_id`, with nodes keyed `<schema>:<id>`. `rs-benthic-v1` is the functional schema (L0 biotic/abiotic, L1 phyla, L2 the 14 MariMap codes, L3 genera), `rs-taxa-v1` the identity schema (`A<AphiaID>` nodes, AphiaID also shipped as `worms_aphia_id`) and `rs-fauna-v1` the fish and fauna schema. Node ids are namespaced in taxonomy 3.0.0: an `rs-taxa-v1` id is `A` followed by the WoRMS AphiaID, and the non-taxon nodes of `rs-benthic-v1` and `rs-fauna-v1` (debris, equipment, divers, wrecks, seafloor, unknown) carry `NT_*` ids. `label-map` maps each source's raw labels to these nodes, one row per source and native label or pixel value. `match_type` is exact, broader, narrower, related or unmapped; `narrower` is stored for audit and never used as a positive. A native label with no mapping is a row with `match_type = unmapped` and null node columns, never dropped, so the unmapped labels of each source are listed in `label-map` (the mapped share per source is tabulated below); the training views send them to `255` (ignore).

### Label overlap

Taxonomy nodes that more than one source reaches. Cross-source training should use these, or the coarse training views.

| Taxonomy node | Sources | Reached by |
|---|---|---|
| `A146419` | 4 | `roboflow-aquarium`, `uiis`, `uiis10k`, `usis10k` |
| `HC` | 4 | `coralscapes`, `reef-support-benthic-own`, `reef-support-seaview-labels`, `seaview-survey-imagery` |
| `MIL` | 4 | `coralscapes`, `reef-support-benthic-own`, `reef-support-seaview-labels`, `seaview-survey-imagery` |
| `A1065` | 3 | `uiis`, `uiis10k`, `usis10k` |
| `A1267` | 3 | `uiis`, `uiis10k`, `usis10k` |
| `A1838` | 3 | `uiis`, `uiis10k`, `usis10k` |
| `A3` | 3 | `uiis`, `uiis10k`, `usis10k` |
| `A51` | 3 | `uiis`, `uiis10k`, `usis10k` |
| `FISH` | 3 | `coralscapes`, `seaview-survey-imagery`, `suim` |
| `NT_DEBRIS` | 3 | `uiis`, `uiis10k`, `usis10k` |
| `NT_DIVER` | 3 | `uiis`, `uiis10k`, `usis10k` |
| `NT_EQUIPMENT` | 3 | `uiis`, `uiis10k`, `usis10k` |
| `NT_REEF` | 3 | `uiis`, `uiis10k`, `usis10k` |
| `NT_SEAFLOOR` | 3 | `uiis`, `uiis10k`, `usis10k` |
| `NT_WRECK` | 3 | `uiis`, `uiis10k`, `usis10k` |
| `OT` | 3 | `coralscapes`, `reef-support-benthic-own`, `reef-support-seaview-labels` |
| `SC` | 3 | `reef-support-benthic-own`, `reef-support-seaview-labels`, `seaview-survey-imagery` |
| `UNKNOWN` | 3 | `roboflow-coral-bleaching-final-v6i`, `roboflow-coral-bleaching-general-v1-yolov8s`, `roboflow-coral-classification-copy-changed-v13i` |
| `WC` | 3 | `coralscapes`, `seaview-survey-imagery`, `suim` |
| `COTS` | 2 | `coralscapes`, `seaview-survey-imagery` |
| `DIV` | 2 | `coralscapes`, `suim` |
| `HC_ACROPORA` | 2 | `coralscapes`, `seaview-survey-imagery` |
| `HC_POCILLOPORA` | 2 | `coralscapes`, `seaview-survey-imagery` |
| `HC_STYLOPHORA` | 2 | `coralscapes`, `seaview-survey-imagery` |
| `RB` | 2 | `coralscapes`, `seaview-survey-imagery` |
| `SCL` | 2 | `reef-support-benthic-own`, `reef-support-seaview-labels` |
| `SD` | 2 | `coralscapes`, `seaview-survey-imagery` |
| `SG` | 2 | `coralscapes`, `seaview-survey-imagery` |
| `TRASH` | 2 | `coralscapes`, `seaview-survey-imagery` |
| `URCH` | 2 | `coralscapes`, `seaview-survey-imagery` |

49 node(s) are reached by a single source.

### Mapped share per source

| Source | Labels | Mapped to the taxonomy |
|---|---|---|
| `coralscapes` | 0 | – |
| `noaa-pifsc-bleaching` | 10,326 | 0.0 % |
| `reef-support-benthic-own` | 0 | – |
| `reef-support-seaview-labels` | 0 | – |
| `roboflow-aquarium` | 4,815 | 100.0 % |
| `roboflow-coral-bleaching-final-v6i` | 3,608 | 0.0 % |
| `roboflow-coral-bleaching-general-v1-yolov8s` | 3,574 | 0.0 % |
| `roboflow-coral-classification-copy-changed-v13i` | 4,189 | 0.0 % |
| `roboflow-coral-reef-classification-v3i` | 4,496 | 0.0 % |
| `seaview-survey-imagery` | 858,320 | 99.9 % |
| `suim` | 0 | – |
| `uiis` | 56,820 | 100.0 % |
| `uiis10k` | 83,722 | 100.0 % |
| `usis10k` | 37,746 | 100.0 % |

## Splits

The card publishes the **measured** split shares of every config, never a nominal ratio: upstream splits dominate several sources, so the v1.0 shares are not 80/10/10 or 70/15/15. New rows target 70/15/15 with the split names `train`, `validation`, `test` and `test_ood`. Precedence, first match wins: (1) an image that matches a v1.0 image by sha256, or within Hamming distance 4 of its dhash, keeps that v1.0 split unless an upstream benchmark manifest pins it (see below), and an image within 5-8 of a v1.0 image is never used across the train/eval boundary; (2) the out-of-distribution holdout rules (deepfish, temperate-australasia, mediterranean, a data-driven tropical province, depth of 800 m or more, platform AUV, year 2025 or later) go to `test_ood`, which is never mixed into `test`; (3) the upstream split where defined; (4) a group hash on survey, site or quadrat, else the near-duplicate group. Images of one group share a split. Group keys in `SPLIT_MAP.json` are namespaced `<source>/<id>`, so equal ids in two sources are two groups.

Version 1.1.0 inherits the v1.0 splits by `image_sha256` from the published v1.0 tree, with one exception: six benchmark sources (`coralscapes`, `suim`, `uiis`, `uiis10k`, `usis10k` and one benchmark not distributed in the open release) are pinned to their upstream manifests. A pin routes the manifest's evaluation split to `test` and a second held-out split, where the source has one, to `validation`. A pin wins over an inherited v1.0 split; the out-of-distribution holdout and any rule that keeps an image out of evaluation win over a pin. The pins moved images that were in v1.0 `train` into the v1.1.0 `validation` and `test` splits and resolved near-duplicate groups that straddle v1.0 splits, while the frozen v1.0 configs keep their v1.0 splits. A group that cannot be assigned consistently is withheld from every table and listed under `withheld` in `SPLIT_MAP.json`.

Because several of these sources carry large upstream test sets, the overall split is not the 70/15/15 target: the test share is well above 15 %, and `boxes` and `instances` are the most affected tables.

**Evaluation caveat.** A model trained on the v1.0 configs must not be evaluated on the v1.1.0 `validation` or `test` splits of these six sources, because some of their images were in `train` in v1.0.

**Left out and overridden (v1.1.0).** No group was withheld. The split map also leaves out 215 images that sit in the prior-release near-duplicate band (dHash distance 5 to 8) of v1.0 train and evaluation images, on the side their near-duplicate group did not keep, and 51 images that each collide on the DCT pHash-64 with an image of another split (one image of the pair goes, never a v1.0 image); these 266 images ship in no split of any config. Where a benchmark pin and v1.0 continuity disagree the pin wins, and it resolves the near-duplicate groups that straddle v1.0 splits: the pins overrode the v1.0 split of 4,544 images (`pins.v1_continuity_overridden` in `SPLIT_MAP.json`).

**Ratios.** The 70/15/15 target is checked per source family, over distinct images, not per config. The NOAA bleaching images, the Roboflow Coral Universe forks and the Seaview sources measure 70.8 / 14.6 / 14.6 %, 68.5 / 15.7 / 15.8 % and 67.6 / 12.6 / 19.8 % (train / validation / test); pinned benchmark sources and strata under 100 images (`roboflow-aquarium`) are left out of the check. Row shares of a config therefore deviate, as the table shows: `points` rows are 65.2 / 12.7 / 22.1 % (its 11,356 distinct images 67.1 / 12.8 / 20.2 %) and `semseg` rows 74.1 / 8.9 / 16.9 %. `noaa-pifsc-bleaching` is assigned by whole site: its 236 sites went 42 to train, 97 to validation and 97 to test, so its image shares are not its site shares.

| Config | train | validation | test | test_ood | Total |
|---|---|---|---|---|---|
| `images` | 22,193 (68.9 %) | 4,562 (14.2 %) | 5,445 (16.9 %) | – | 32,200 |
| `metadata` | 40,183 (61.5 %) | 8,096 (12.4 %) | 17,077 (26.1 %) | – | 65,356 |
| `points` | 559,490 (65.2 %) | 108,730 (12.7 %) | 190,100 (22.1 %) | – | 858,320 |
| `bleaching` | 19,631 (71.8 %) | 3,823 (14.0 %) | 3,889 (14.2 %) | – | 27,343 |
| `boxes` | 49,038 (52.2 %) | 8,417 (9.0 %) | 36,503 (38.9 %) | – | 93,958 |
| `semseg` | 2,445 (74.1 %) | 295 (8.9 %) | 559 (16.9 %) | – | 3,299 |
| `instances` | 45,869 (51.5 %) | 7,664 (8.6 %) | 35,612 (39.9 %) | – | 89,145 |
| `benthic-cover` | 3,298 (100.0 %) | – | – | – | 3,298 |
| `taxonomy` | 2,937 (100.0 %) | – | – | – | 2,937 |
| `label-map` | 306 (100.0 %) | – | – | – | 306 |
| `sources` | 14 (100.0 %) | – | – | – | 14 |

## Dataset creation

### Sources

| Source | Version | Rights holder | Verified | Provenance | Images | Configs |
|---|---|---|---|---|---|---|
| `coralscapes` | 1.0 | – | dataset-card 2026-08-18 | https://josauder.github.io/coralscapes/ | 2,074 | benthic-cover,metadata,semseg |
| `noaa-pifsc-bleaching` | 1-image-labels | – | dataset-card 2026-08-17 | – | 10,326 | bleaching,images,metadata |
| `reef-support-benthic-own` | 2026-10-06 | – | licence-file 2026-08-17 | – | 1,226 | benthic-cover,metadata,semseg |
| `reef-support-seaview-labels` | 2026-10-06 | – | licence-file 2026-10-06 | – | 2,707 | images,metadata |
| `roboflow-aquarium` | zip-5d30fed3dd5a | – | licence-file 2026-10-06 | https://public.roboflow.ai/object-detection/aquarium | 638 | boxes,images,metadata |
| `roboflow-coral-bleaching-final-v6i` | v6i-image-labels-r2 | – | licence-file 2026-09-24 | – | 2,518 | bleaching,images,metadata |
| `roboflow-coral-bleaching-general-v1-yolov8s` | v1-yolov8s-image-labels-r2 | – | licence-file 2026-09-24 | – | 1,899 | bleaching,images,metadata |
| `roboflow-coral-classification-copy-changed-v13i` | v13i-image-labels | – | licence-file 2026-09-24 | – | 1,167 | bleaching,images,metadata |
| `roboflow-coral-reef-classification-v3i` | v3i-image-labels | – | licence-file 2026-09-24 | – | 4,409 | bleaching,images,metadata |
| `seaview-survey-imagery` | 2026-10-06 | – | licence-file 2026-09-24 | https://espace.library.uq.edu.au/view/UQ:734799 | 11,869 | images,metadata,points |
| `suim` | 2020 | – | licence-file 2026-08-29 | – | 1,585 | metadata |
| `uiis` | rev-44ca5db46599 | – | dataset-card 2026-10-06 | https://huggingface.co/datasets/LiamLian0727/UIIS | 4,285 | boxes,instances,metadata |
| `uiis10k` | rev-5c4316368a8e | – | dataset-card 2026-10-06 | https://huggingface.co/datasets/LiamLian0727/UIIS10K | 10,042 | boxes,instances,metadata |
| `usis10k` | rev-b10f41ab819b | – | dataset-card 2026-10-06 | https://huggingface.co/datasets/LiamLian0727/USIS10K | 10,611 | boxes,instances,metadata |

### Build and gates

Built with the [marine-data](https://github.com/reefsupport/marine-data) registry; every source is admitted only after its licence is recorded, and the licence gate raises, never warns. The release is pinned by `provenance/release.lock.yaml`. Gate results are published as `provenance/GATES.json`. Near-duplicate images inside one split that come from different Roboflow Universe forks of one image set, or from the Seaview survey imagery and our labelled copies of its photographs, are disclosed, not removed: the dedup gate counts them as `near_pairs_across_sources_in_family` in `provenance/GATES.json`, and a near-duplicate across splits still fails it. In v1.1.0, 5,201 such pairs are disclosed. A further 6 pairs from different sources fall within the pHash-64 threshold (8 bits) but are kept: the pHash-256 of the two images differs by more than 32 bits and both carry at least 4 bits of grey entropy, so the gate waives them as look-alikes and counts them as `near_pairs_across_sources_unconfirmed`.

## Personal and sensitive information

The images are underwater and reef photographs. Divers, and rarely faces, can appear. The release gates include a privacy pass (face detections at a score of 0.85 or more are excluded, 0.60 to 0.85 are flagged) and a check for EXIF GPS; their results are published in `provenance/GATES.json`. Images the detector flags are reviewed on contact sheets: an image with an identifiable person is withheld from every table, and a detection that proves to be a false positive is recorded in `registry/privacy/review.yaml` in the code repository. No image with an identifiable person is published knowingly; one Roboflow fork was excluded from this release for that reason (see Excluded sources). A gate and a review lower the risk; they do not remove it. The EXIF GPS of `seaview-survey-imagery` images is retained by design, because that source publishes per-image coordinates. `metadata` carries position at the precision the source provides (`coord_precision_m`) and no source is currently flagged location-sensitive in the registry. In v1.1.0 the privacy gate scanned 32,200 images: 10 flagged images were reviewed by hand and cleared, none was withheld and no unreviewed detection remains. Divers are an annotated class, kept on purpose: `coralscapes`, `uiis`, `uiis10k` and `usis10k` label them (`human` and `human divers`, 7 `label-map` rows). No face or person identification is intended or supported. To drop the class, filter out the rows whose `taxon_node_id` ends in `NT_DIVER` (`label-map` lists the labels that map to it) and send its value to `255` (ignore) in the masks.

## Limitations and known defects

- Coralscapes has no soft-coral or octocoral class. `label-map` records the absent classes, so a model trained on Coralscapes alone cannot predict them and cover for them is undercounted.
- Absence is not a negative. For the reef-support and Seaview label sources `255` means unannotated, not background; Coralscapes uses its own ignore id. `label-map` carries the per-source ignore semantics.
- Cross-source benthic training should use the coarse training views, or mask the classes a source cannot contain: the label overlap table shows how few classes more than one source reaches.
- Labels without a taxonomy mapping are kept (`mapped = false`) and ignored in the training views. A source whose mapped share is below the 95 % floor is listed in the sources table with its share.
- Until the declared bridges from `rs-taxa-v1` to `rs-benthic-v1` and `rs-fauna-v1` exist, rows labelled by an identity node carry a null `rs_benthic_code`.
- A small number of identical images carry conflicting labels; every party has `label_status = conflict` and all rows are kept. The sampled label-quality audit has not been run.
- Confidence is present only where the upstream source supplies it; forks and crowd sources have none. It is never invented.
- v1.0 files are frozen as published, with their known defects: Seaview rows say `CC-BY-3.0-AU` where the University of Queensland licence is CC BY 3.0 unported (new rows and `metadata.license` say `CC-BY-3.0`); v1.0 `fish-boxes` uses `license`, `source_id` and int32 where the other configs use `licence`, `source` and int64; v1.0 masks render black in the viewer because they hold class ids, not colours (new semantic masks carry a palette).
- Coordinates, dates and depths are as complete as each source made them; analytics resolvers exist for a subset of the sources.
- The images of `seaview-survey-imagery` are in `images` and `metadata`, and its point labels are in `points` only; no Seaview row is part of `benthic-cover`: the cover rollup reads the original point sources, and Seaview points stay out of it until a cover decision is made for that source.
- **Licensing and provenance caveats.** Four sources are forks of Roboflow Universe projects (`roboflow-coral-bleaching-final-v6i`, `roboflow-coral-bleaching-general-v1-yolov8s`, `roboflow-coral-classification-copy-changed-v13i`, `roboflow-coral-reef-classification-v3i`). They are web-scraped photo sets re-shared on Roboflow. Their CC-BY-4.0 tag is the uploader's claim, which Reef Support could not verify with the original photographers, and third-party watermarks appear sporadically. Near-duplicate images across these four sources are reported by the dedup gate and are not removed. A fifth Roboflow fork is excluded from this release: its images are augmented broadcast-video frames that show identifiable people, and its licence cannot be verified. Rights holders can use the takedown route below.
- 24 hidden-test images that v1.0 does not already publish are withheld from this release: they appear in none of its configs. The 1,226 hidden-test images that v1.0 already published are not withheld: they stay in v1.0 and in the new tables.
- Sources below the 95 % mapped floor: `noaa-pifsc-bleaching` (0.0 %), `roboflow-coral-bleaching-final-v6i` (0.0 %), `roboflow-coral-bleaching-general-v1-yolov8s` (0.0 %), `roboflow-coral-classification-copy-changed-v13i` (0.0 %), `roboflow-coral-reef-classification-v3i` (0.0 %).

## Licensing

The data are not under one licence. Every row records its `licence`, `annotation_licence` and `attribution`; the matrix lists each source. Combine sources only in ways each licence allows. `LICENSE` and `NOTICE` list the same notices. The table records the licences as Reef Support read them from the primary sources and is not legal advice.

| Source | Image licence | Annotation licence | Attribution | Class |
|---|---|---|---|---|
| `coralscapes` | [Apache-2.0](https://www.apache.org/licenses/LICENSE-2.0) | Apache-2.0 | Coralscapes (Apache-2.0) | open |
| `noaa-pifsc-bleaching` | US-GOV-PD | US-GOV-PD | NOAA PIFSC-ESD coral bleaching dataset (US-GOV-PD) | open |
| `reef-support-benthic-own` | [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/) | CC-BY-4.0 | Reef Support benthic instance masks — own imagery (Colombia) (CC-BY-4.0) | open |
| `reef-support-seaview-labels` | [CC BY 3.0](https://creativecommons.org/licenses/by/3.0/) | CC-BY-3.0 | Reef Support annotations on XL Catlin Seaview imagery (CC-BY-3.0) | open |
| `roboflow-aquarium` | [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/) | CC-BY-4.0 | Roboflow Aquarium Combined (CC-BY-4.0) | open |
| `roboflow-coral-bleaching-final-v6i` | [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/) | CC-BY-4.0 | Coral Bleaching Final (Roboflow v6i) (CC-BY-4.0) | open |
| `roboflow-coral-bleaching-general-v1-yolov8s` | [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/) | CC-BY-4.0 | Coral Bleaching_General (Roboflow v1, YOLOv8s) (CC-BY-4.0) | open |
| `roboflow-coral-classification-copy-changed-v13i` | [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/) | CC-BY-4.0 | Coral Classification Copy -Changed- (Roboflow v13i) (CC-BY-4.0) | open |
| `roboflow-coral-reef-classification-v3i` | [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/) | CC-BY-4.0 | Coral Reef Classification (Roboflow v3i) (CC-BY-4.0) | open |
| `seaview-survey-imagery` | [CC BY 3.0](https://creativecommons.org/licenses/by/3.0/) | CC-BY-3.0 | SEAVIEW raw survey imagery (UQ eSpace) (CC-BY-3.0) | open |
| `suim` | MIT | MIT | SUIM (Semantic Segmentation of Underwater Imagery) (MIT) | open |
| `uiis` | [Apache-2.0](https://www.apache.org/licenses/LICENSE-2.0) | Apache-2.0 | UIIS (WaterMask) underwater instance segmentation (Apache-2.0) | open |
| `uiis10k` | [Apache-2.0](https://www.apache.org/licenses/LICENSE-2.0) | Apache-2.0 | UIIS10K underwater instance segmentation (UWSAM) (Apache-2.0) | open |
| `usis10k` | [Apache-2.0](https://www.apache.org/licenses/LICENSE-2.0) | Apache-2.0 | USIS10K underwater salient instance segmentation (Apache-2.0) | open |

Credits: Marine Ecoregions of the World (MEOW): Spalding, M. D. et al. (2007), BioScience 57(7), 573-583. Only the derived realm, province and ecoregion names ship; the MEOW geometry does not. World Register of Marine Species (WoRMS Editorial Board), https://www.marinespecies.org, CC BY 4.0. AphiaIDs identify taxa in the taxonomy, `label-map` and the label tables.

## Excluded sources

Every registered source that is not in this release falls under one of the reasons below. The sources that are in it are listed in the `sources` config.

- **restricted-nd**: no derivatives: a verbatim-only licence cannot cover cropped, resized or re-labelled releases
- **internal-only / unknown**: no licence granting redistribution was established for these sources (research-use-only or nothing on record)
- **per-row licence**: sources whose every item carries its own licence are held out until each row's licence is resolved
- **A fifth Roboflow fork**: tagged release-excluded. Its images are augmented broadcast-video frames that show identifiable people, and its licence tag comes from the uploader's sidecar and cannot be verified
- **A sixth Roboflow fork**: dropped as a duplicate of an image set already in this release; none of its images is included
- **A fish-market photo source**: tagged release-excluded and quarantined: out-of-water market photographs of fish
- **sonar sources**: side-scan and forward-looking sonar imagery is a different sensing modality from the camera images of this release; none is included
- **open-licence sources not yet staged**: registered sources with an open licence that were not in the v1.1.0 fetch list (for example `noaa-pifsc-esa-coral-icra`, `marineevt` and `salmon-cage`) are deferred to a later release
- **non-image sources**: text-only sources such as `oceaninstruct` hold no images
- **One non-commercial source**: licensed for non-commercial use only; it ships in the separate non-commercial flavour of the release, not in this repository. No other non-commercial source is part of v1.1.0

## Takedown

Rights holders and data subjects: open an issue at https://github.com/reefsupport/marine-data/issues. Accepted requests are published as a new version.

## Versions and changelog

| Version | Date | Notes |
|---|---|---|
| v1.0 | 2026-10-07 | coral-masks, scene-masks, instance-masks and fish-boxes: 8 sources, 58,803 rows. Frozen: files, rows and paths are unchanged in v1.1.0. |
| v1.1.0 | 2026-10-17 | Additive: 11 new configurations. |
| v1.1.1 | 2026-10-24 | Additive: one new configuration, `condition-masks` (635 Tayrona bleaching rasters, `train`). Every v1.1.0 file is unchanged. |

- **Added (additive; no v1.0 file, row, path or `configs:` entry changed):** the configs `images`, `metadata`, `points`, `bleaching`, `boxes`, `semseg`, `instances`, `benthic-cover`, `sites`, `label-map` and `sources`, and `taxonomy` in the open repo; `LICENSE`, `NOTICE`, `CITATION.cff`, `croissant.json` and `provenance/`.
- **Unchanged:** every v1.0 config, with its files, rows and paths; the release is additive.
- **Task layers:** the v2 task layers are new tables on the `image_sha256` key: `metadata`, `points`, `bleaching`, `boxes`, `semseg` and `instances` (`benthic-cover` and `sites` are derived from them).
- **Changed:** this card (text only; the v1.0 `configs:` entries are carried verbatim). `dataset_info` is written for the new configs only.
- **Label space:** taxonomy 3.0.0. Every new table row carries `taxonomy_version`; the Coralscapes class "unknown hard substrate" is unmapped in the new tables; three wrong AphiaIDs were corrected (205902, 206485, 758259). v1.0 `class_map` columns are untouched.
- **Splits:** a split map that inherits the v1.0 splits by `image_sha256` (v1.0 continuity); six benchmark sources are pinned to their upstream manifests instead, which moved some images out of v1.0 `train` into `validation` and `test`; the frozen v1.0 configs keep their v1.0 splits (see Splits).
- **Excluded:** a fifth Roboflow fork, and a sixth dropped as a duplicate (see Excluded sources).
- **Licences:** new Seaview rows and `metadata.license` use `CC-BY-3.0` (v1.0 rows keep `CC-BY-3.0-AU`, declared above).

Files of a released version are not edited. Pin a version with `revision`:

```python
load_dataset("reefsupport/marine-data", "metadata", split="train", revision="v1.1.0")
```

## Citation

```bibtex
@misc{reef_support_marine_data_2026,
  title = {marine-data},
  author = {Reef Support},
  year = {2026},
  version = {1.1.1},
  url = {https://huggingface.co/datasets/reefsupport/marine-data},
}
```

Please also cite the sources you use:

- `coralscapes`: Sauder et al., The Coralscapes Dataset, ICCV 2025 Marine Vision Workshop
- `noaa-pifsc-bleaching`: Pacific Islands Fisheries Science Center (2025). Ecosystem Sciences Division (ESD). NOAA-ESD-CORAL-Bleaching-Dataset v1. https://huggingface.co/datasets/NMFS-OSI/NOAA-PIFSC-ESD-CORAL-Bleaching-Dataset
- `reef-support-seaview-labels`: González-Rivero et al. (2019). Seaview Survey Photo-quadrat and Image Classification Dataset. University of Queensland. doi:10.14264/UQL.2019.930 Masks and labels: Reef Support annotations (CC-BY-4.0).
- `roboflow-aquarium`: Roboflow 2020, Aquarium Combined (public.roboflow.ai/object-detection/aquarium)
- `roboflow-coral-bleaching-final-v6i`: Lockie. "Coral Bleaching Final" (Roboflow Universe, v6i). https://universe.roboflow.com/lockie/coral-bleaching-final
- `roboflow-coral-bleaching-general-v1-yolov8s`: Lockie. "Coral Bleaching_General" (Roboflow Universe, v1, YOLOv8s). https://universe.roboflow.com/lockie/coral-bleaching_general/dataset/1
- `roboflow-coral-classification-copy-changed-v13i`: maxyn-icaonapo. "Coral Classification Copy (Changed)" (Roboflow Universe, v13). https://universe.roboflow.com/maxyn-icaonapo/coral-classification-copy-changed
- `roboflow-coral-reef-classification-v3i`: twork. "Coral Reef Classification" (Roboflow Universe). https://universe.roboflow.com/twork/coral-reef-classification
- `seaview-survey-imagery`: González-Rivero et al. (2019). Seaview Survey Photo-quadrat and Image Classification Dataset. University of Queensland. doi:10.14264/UQL.2019.930
- `uiis`: Lian et al. 2023, WaterMask: Instance Segmentation for Underwater Imagery, ICCV 2023, pp.1305-1315
- `uiis10k`: Li et al. 2025, Advancing Marine Research: UWSAM Framework and UIIS10K Dataset for Precise Underwater Instance Segmentation, arXiv:2505.15581
- `usis10k`: Lian et al. 2024, Diving into Underwater: Segment Anything Model Guided Underwater Salient Instance Segmentation and A Large-scale Dataset, ICML 2024 (PMLR v235:29545-29559)

## Contact

hello@reef.support