{"success":true,"data":{"author":"Sauder J., Tuia D. et al. (EPFL ECEO)","bucketAccessType":"public","bucketName":"josauder/coralscapesV2","bucketPrefix":"","category":"Environmental Science","cloudProvider":"huggingface","createdAt":"2026-11-16T02:38:32.488Z","description":"CoralscapesV2 is a scene-understanding dataset for coral reefs and the first for panoptic segmentation underwater (ECCV 2026 Workshop on Marine Vision, arXiv 2609.12826), extending Coralscapes (EPFL ECEO). It holds 2,433 densely annotated video frames at 1024x2048 px from 41 dive sites in the Red Sea region, split by site into 1,790 train, 170 validation and 473 test images. Semantic masks use 95 fine-grained benthic and visual categories (live, bleached and dead hard and soft coral genera, seagrass, algae, sand, rock, trash, fish and others), each mapped onto the 39 original Coralscapes classes so older models still evaluate. An instances configuration adds 65,738 exhaustive fish instance masks annotated with video context, and a sequences configuration returns 31-frame clips around each labelled anchor frame for temporal models. Measured: 524 Parquet and metadata files, about 220 GB. Load with datasets.load_dataset(\"josauder/coralscapesV2\"). Licensed Apache 2.0; cite CoralscapesV2 and Coralscapes (Sauder et al.).","displayTitle":"CoralscapesV2: Panoptic Segmentation of Red Sea Coral Reefs with 95 Benthic Classes and Fish Instances","id":"530858a6-509b-5f81-92e5-7a0b9fb5a06e","license":"Apache-2.0","licenseEvidenceUrl":"https://huggingface.co/datasets/josauder/coralscapesV2","objectCountEstimate":524,"probeStatus":"ok","referenceUrl":"https://arxiv.org/abs/2609.12826","region":"us","schemaCard":{"accessHints":[{"language":"python-datasets","snippet":"from datasets import load_dataset\nds = load_dataset(\"josauder/coralscapesV2\")"},{"language":"python-hub","snippet":"from huggingface_hub import snapshot_download\nsnapshot_download(repo_id=\"josauder/coralscapesV2\", repo_type=\"dataset\")"}],"caveats":["HuggingFace metadata: author=josauder, license:apache-2.0, size_categories:1K<n<10K, format:parquet, downloads=8044"],"format":"mixed","generatedAt":"2026-11-16T02:38:37.356Z","layout":{"example":"data/test-00000-of-00003.parquet","pattern":"{repoId}/{filename}"},"mountPath":"/mnt/archive","sampleHeaders":[{"bytes":"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","encoding":"hex","key":"data/test-00000-of-00003.parquet","truncatedAt":512}],"sampleObjects":["data/test-00000-of-00003.parquet","data/test-00001-of-00003.parquet","data/test-00002-of-00003.parquet","data/train-00000-of-00009.parquet","data/train-00001-of-00009.parquet"],"sidecars":[{"bytes":"---\ndataset_info:\n- config_name: default\n  features:\n  - name: image\n    dtype: image\n  - name: label\n    dtype: image\n  splits:\n  - name: train\n    num_bytes: 4953203364.0\n    num_examples: 1790\n  - name: validation\n    num_bytes: 412061003.0\n    num_examples: 170\n  - name: test\n    num_bytes: 1386298543.0\n    num_examples: 473\n  download_size: 6751562910.0\n  dataset_size: 6751562910.0\n- config_name: instances\n  features:\n  - name: image\n    dtype: image\n  - name: label\n    dtype: image\n  - name: instance_ids\n    dtype: image\n  - name: stem\n    dtype: string\n  - name: site\n    dtype: string\n  splits:\n  - name: train\n    num_bytes: 5046081675.0\n    num_examples: 1790\n  - name: validation\n    num_bytes: 420185675.0\n    num_examples: 170\n  - name: test\n    num_bytes: 1410234815.0\n    num_examples: 473\n  download_size: 6876502165.0\n  dataset_size: 6876502165.0\n- config_name: sequences\n  features:\n  - name: frames\n    sequence: image\n  - name: frame_offsets\n    sequence: int16\n  - name: source_frame_offsets\n    sequence: int16\n  - name: anchor_index\n    dtype: int8\n  - name: label\n    dtype: image\n  - name: stem\n    dtype: string\n  - name: site\n    dtype: string\n  splits:\n  - name: train\n    num_bytes: 151218608718.0\n    num_examples: 1790\n  - name: validation\n    num_bytes: 12588004132.0\n    num_examples: 170\n  - name: test\n    num_bytes: 42486693602.0\n    num_examples: 473\n  download_size: 206293306452.0\n  dataset_size: 206293306452.0\nconfigs:\n- config_name: default\n  default: true\n  data_files:\n  - split: train\n    path: data/train-*\n  - split: validation\n    path: data/validation-*\n  - split: test\n    path: data/test-*\n- config_name: instances\n  data_files:\n  - split: train\n    path: instances/train-*\n  - split: validation\n    path: instances/validation-*\n  - split: test\n    path: instances/test-*\n- config_name: sequences\n  data_files:\n  - split: train\n    path: sequences/train-*\n  - split: validation\n    path: sequences/validation-*\n  - split: test\n    path: sequences/test-*\ntask_categories:\n- image-segmentation\ntags:\n- biology\n- ecology\n- climate\nlicense: apache-2.0\nsize_categories:\n- 1K<n<10K\n---\n\n## Optional configurations\n\nThe default configuration remains the original `image` + `label` dataset and is loaded with `load_dataset(REPO)`. Optional data is stored in separate Parquet shards and is downloaded only when its configuration is requested.\n\n### Instances\n\n`load_dataset(REPO, \"instances\")` adds a uint16 `instance_ids` image. Fish is class 25 and per-fish IDs are `25 * 2000 + instance_index`; a binary mask is `instance_ids == instance_id`.\n\n### Sequences\n\n`load_dataset(REPO, \"sequences\", streaming=True)` returns `frames` ordered from offset -20 through +10, with the labeled anchor at `anchor_index == 20`. At a genuine video boundary, unavailable frames repeat the nearest available frame and `source_frame_offsets` records that padding.\n\n## CoralscapesV2\n\nPaper: [CoralscapesV2: Panoptic and Fine-Grained Visual Scene Understanding in Coral Reefs](https://arxiv.org/abs/2609.12826) (ECCV 2026 Workshop on Marine Vision) &middot; Project page: [josauder.github.io/coralscapesv2](https://josauder.github.io/coralscapesv2/)\n\nCoralscapesV2 is a general-purpose scene-understanding dataset for coral reefs and the first dataset for panoptic segmentation in reefs. It extends the original [Coralscapes](https://huggingface.co/datasets/EPFL-ECEO/coralscapes) dataset along every measurable axis: more annotated frames, higher annotation density and quality, a refined label set of **95 fine-grained visual categories** (up from 39), and **65,738 exhaustive fish instance masks** annotated using video context.\n\n## Dataset Structure\nCoralscapesV2 spans 2433 densely-annotated video frames at 1024×2048px resolution gathered from 41 dive sites in the Red Sea region, labeled in a consistent and speculation-free manner over **95 fine-grained visual categories**.\n\nThe released semantic label masks (`label`) use the 95-class fine-grained taxonomy: class ids are given by `id2label.json` and colors by `label2color.json`. Every fine class also maps","encoding":"utf8","key":"README.md","name":"README.md","size":14582,"truncated":true}],"sourceObjectKey":"data/test-00000-of-00003.parquet","starterCell":"from datasets import load_dataset\nds = load_dataset(\"josauder/coralscapesV2\")\nprint(ds)","starterSamplePath":"data/test-00000-of-00003.parquet","totalObjects":524},"summary":"CoralscapesV2 is a scene-understanding dataset for coral reefs and the first for panoptic segmentation underwater (ECCV 2026 Workshop on Marine Vision, arXiv 2609.12826),…","title":"CoralscapesV2: Panoptic Segmentation of Red Sea Coral Reefs with 95 Benthic Classes and Fish Instances","updatedAt":"2026-11-16T17:02:44.621Z","visibility":"public","websiteUrl":"https://josauder.github.io/coralscapesv2/"}}