{"success":true,"data":{"author":"Chaoqun Cheng (source data); Paninski Lab, Columbia University (packaging)","bucketAccessType":"public","bucketName":"paninski-lab/marmoset3k","bucketPrefix":"","category":"Neuroscience","cloudProvider":"huggingface","createdAt":"2026-11-21T20:45:58.794Z","description":"A labeled pose estimation dataset for the common marmoset (Callithrix jacchus), a primate model in neuroscience and social behavior research, recorded from a top-down camera with animals alone or in pairs. In pair sessions each labeled frame is a crop of one focal animal, and 16 full-body keypoints are labeled per animal. In-distribution and out-of-distribution label sets contain different animals. Repackaged by the Paninski Lab at Columbia University for Lightning Pose: labeled-data/ PNG frames with neighboring context frames, CollectedData.csv and CollectedData_test.csv (DeepLabCut-style 2D keypoint labels), in-distribution videos, a training config (256 by 256 resize, ResNet-50 AP-10K backbone) and project.yaml; about 0.3 GB. Suited to benchmarking markerless primate pose tracking, semi-supervised pose estimation and multi-animal tracking. Source data: Marmoset3K, Zenodo record by Chaoqun Cheng, DOI 10.5281/zenodo.14672425, CC BY 4.0. Cite that record and Lightning Pose (Biderman et al., Nature Methods 2024).","displayTitle":"Marmoset3k: Top-Down Pose Estimation Keypoints for Common Marmosets (Callithrix jacchus), Lightning Pose Format (Columbia University Paninski Lab)","doi":"10.5281/zenodo.14672425","huggingFaceUrl":"https://huggingface.co/datasets/paninski-lab/marmoset3k","id":"355d4379-2351-54d4-b38f-12b9d066818f","license":"CC-BY-4.0","licenseEvidenceUrl":"https://zenodo.org/api/records/14672425","objectCountEstimate":4750,"probeStatus":"ok","referenceUrl":"https://huggingface.co/datasets/paninski-lab/marmoset3k","region":"us","schemaCard":{"accessHints":[{"language":"python-datasets","snippet":"from datasets import load_dataset\nds = load_dataset(\"paninski-lab/marmoset3k\")"},{"language":"python-hub","snippet":"from huggingface_hub import snapshot_download\nsnapshot_download(repo_id=\"paninski-lab/marmoset3k\", repo_type=\"dataset\")"}],"caveats":["HuggingFace metadata: author=paninski-lab, license:cc-by-4.0, size_categories:1K<n<10K, format:imagefolder, downloads=399"],"format":"mixed","generatedAt":"2026-11-21T20:48:47.778Z","layout":{"example":"CollectedData.csv","pattern":"{repoId}/{filename}"},"mountPath":"/mnt/archive","sampleHeaders":[{"bytes":"scorer,human,human,human,human,human,human,human,human,human,human,human,human,human,human,human,human,human,human,human,human,human,human,human,human,human,human,human,human,human,human,human,human\nbodyparts,head,head,leftear,leftear,rightear,rightear,neck,neck,leftelbow,leftelbow,rightelbow,rightelbow,lefthand,lefthand,righthand,righthand,spinemid,spinemid,tailbase,tailbase,leftknee,leftknee,rightknee,rightknee,leftfoot,leftfoot,rightfoot,rightfoot,tailmid,tailmid,tailend,tailend\ncoords,x,y,x,y,x,y,x,y,x,","encoding":"utf8","key":"CollectedData.csv","truncatedAt":512}],"sampleObjects":["CollectedData.csv","CollectedData_test.csv","config_marmoset3k.yaml","labeled-data/pair_test_01/img00000001.png","labeled-data/pair_test_01/img00000002.png"],"sidecars":[{"bytes":"---\ntask_categories:\n  - keypoint-detection\nlicense: cc-by-4.0\ntags:\n  - biology\n  - pose-estimation\n  - marmoset\n  - primate\n  - lightning-pose\npretty_name: Marmoset3k\nsize_categories:\n  - 1K<n<10K\n---\n\n# Marmoset3k — Lightning Pose Single-View Dataset\n\nTop-down pose estimation dataset for common marmosets (*Callithrix jacchus*), packaged for use with [Lightning Pose](https://github.com/paninski-lab/lightning-pose).\n\n## Dataset Description\n\nMarmosets were recorded from a top-down camera in two configurations: alone (single) and in pairs. For pair sessions, individual animals are tracked — each labeled frame contains one focal animal cropped from a scene with two marmosets. **16 full-body keypoints** are labeled per animal.\n\nSource data: original archive at https://zenodo.org/records/14672425.\n\n## Data Splits\n\n| Split | Labeled frames | Sessions |\n|-------|---------------:|---------:|\n| In-distribution (InD) | 3,920 | 4 (2 pair, 2 single) |\n| Out-of-distribution (OOD) | 824 | 2 (1 pair, 1 single) |\n\nInD and OOD sessions contain **different animals**.\n\n- `CollectedData.csv` — InD labels; `videos/` — InD videos\n- `CollectedData_test.csv` — OOD labels\n\n## Keypoints\n\n16 full-body keypoints.\n\n| Region | Keypoints |\n|--------|-----------|\n| Head | head, leftear, rightear |\n| Torso | neck, spinemid |\n| Forelimbs | leftelbow, rightelbow, lefthand, righthand |\n| Hindlimbs | leftknee, rightknee, leftfoot, rightfoot |\n| Tail | tailbase, tailmid, tailend |\n\n## Directory Structure\n\n```\nmarmoset3k/\n├── labeled-data/              # Extracted frames per session; includes ±2 context frames\n├── videos/                    # InD session video clips\n├── CollectedData.csv          # InD 2D keypoint labels (x,y per keypoint)\n├── CollectedData_test.csv     # OOD 2D keypoint labels\n├── config_marmoset3k.yaml     # Sample Lightning Pose training config\n└── project.yaml               # View and keypoint definitions (required by LP App)\n```\n\nSee the Lightning Pose documentation for full details on the [single-view data directory structure](https://lightning-pose.readthedocs.io/en/latest/source/directory_structure_reference/singleview_structure.html).\n\n## Usage with Lightning Pose\n\nThe included `config_marmoset3k.yaml` is a ready-to-use training config. Key settings:\n\n- **Image resize:** 256 × 256\n- **Backbone:** `resnet50_animal_ap10k`\n- **Keypoints:** 16\n\nUpdate `data.data_dir` to an absolute path on your machine before training.\n\n```bash\nlitpose train config_marmoset3k.yaml\n```\n\n## Citation\n\nIf you use this dataset, please cite:\n\n```bibtex\n@article{cheng2025real,\n  title     = {A real-time, multi-subject three-dimensional pose tracking system\n               for the behavioral analysis of non-human primates},\n  author    = {Cheng, Chaoqun and Huang, Zijian and Zhang, Ruiming and Huang, Guozheng\n               and Wang, Han and Tang, Likai and Wang, Xiaoqin},\n  journal   = {Cell Reports Methods},\n  volume    = {5},\n  number    = {2},\n  year      = {2025},\n  publisher = {Elsevier}\n}\n```\n\nOriginal data archive: https://zenodo.org/records/14672425.\n","encoding":"utf8","key":"README.md","name":"README.md","size":3122,"truncated":false}],"sourceObjectKey":"CollectedData.csv","starterCell":"import pandas as pd\ndf = pd.read_csv(\"hf://datasets/paninski-lab/marmoset3k/CollectedData.csv\")\ndf.head()","starterSamplePath":"CollectedData.csv","totalObjects":4750},"summary":"A labeled pose estimation dataset for the common marmoset (Callithrix jacchus), a primate model in neuroscience and social behavior research, recorded from a top-down camera with…","title":"Marmoset3k: Top-Down Pose Estimation Keypoints for Common Marmosets (Callithrix jacchus), Lightning Pose Format (Columbia University Paninski Lab)","updatedAt":"2026-11-22T01:07:24.241Z","visibility":"public","websiteUrl":"https://github.com/paninski-lab/lightning-pose"}}