paninski-lab/mirror-fish

Mirror Fish — Lightning Pose Single-View Dataset

Single-camera pose estimation dataset for mormyrid fish body keypoints, packaged for use with Lightning Pose.

Dataset Description

Weakly-electric mormyrid fish (Gnathonemus petersii) swim freely in and out of an experimental tank, capturing worms from a well. The tank has a side mirror and a top mirror, both at 45°, allowing a single camera to capture three views simultaneously — a direct view and two mirror views — at 300 Hz. Each frame is labeled with 51 keypoints: 17 body parts across all three views.

Source data: original archive at https://doi.org/10.6084/m9.figshare.24993363. Data collected by Federico Pedraja, David Ehrlich, and Dillon Noone in the Sawtell Lab, Columbia University.

Data Splits

Split	Labeled frames	Sessions
In-distribution (InD)	373	28
Out-of-distribution (OOD)	94	10

InD and OOD sets contain different sessions / animals (no overlap).

• CollectedData.csv — InD labels; videos/ — InD videos
• CollectedData_test.csv — OOD labels; videos_test/ — OOD videos

Keypoints

51 keypoints total: 17 body parts × 3 views (_main, _top, _right).

Body part	Main view	Top view	Right view
Chin tip	chin_tip_main	chin_tip_top	chin_tip_right
Chin ¾	chin3_4_main	chin3_4_top	chin3_4_right
Chin half	chin_half_main	chin_half_top	chin_half_right
Chin ¼	chin1_4_main	chin1_4_top	chin1_4_right
Chin base	chin_base_main	chin_base_top	chin_base_right
Head	head_main	head_top	head_right
Mid	mid_main	mid_top	mid_right
Tail neck	tail_neck_main	tail_neck_top	tail_neck_right
Caudal ventral	caudal_v_main	caudal_v_top	caudal_v_right
Caudal dorsal	caudal_d_main	caudal_d_top	caudal_d_right
Pectoral L base	pectoral_L_base_main	pectoral_L_base_top	pectoral_L_base_right
Pectoral L	pectoral_L_main	pectoral_L_top	pectoral_L_right
Pectoral R base	pectoral_R_base_main	pectoral_R_base_top	pectoral_R_base_right
Pectoral R	pectoral_R_main	pectoral_R_top	pectoral_R_right
Dorsal	dorsal_main	dorsal_top	dorsal_right
Anal	anal_main	anal_top	anal_right
Fork	fork_main	fork_top	fork_right

Directory Structure

mirror-fish/
├── labeled-data/                    # Extracted frames per session; includes ±2 context frames
├── videos/                          # InD session video clips
├── videos_test/                     # OOD session video clips
├── videos-for-each-labeled-frame/   # 51-frame videos centered on each OOD labeled frame
├── CollectedData.csv                # InD 2D keypoint labels (x,y per keypoint)
├── CollectedData_test.csv           # OOD 2D keypoint labels
├── config_mirror-fish.yaml          # Sample Lightning Pose training config
└── project.yaml                     # View and keypoint definitions (required by LP App)

The videos-for-each-labeled-frame/ directory contains 51-frame video clips with the labeled frame at the center, intended for use with temporal smoothers such as the Ensemble Kalman Smoother.

See the Lightning Pose documentation for full details on the single-view data directory structure.

Usage with Lightning Pose

The included config_mirror-fish.yaml is a ready-to-use training config. Key settings:

• Image resize: 256 × 384
• Backbone: resnet50_animal_ap10k
• Keypoints: 51
• Mirror columns: [0–16] (main), [17–33] (top), [34–50] (right)

Update data.data_dir to an absolute path on your machine before training.

litpose train config_mirror-fish.yaml

Citation

If you use this dataset, please cite:

@article{biderman2024lightning,
  title     = {Lightning Pose: improved animal pose estimation via semi-supervised
               learning, Bayesian ensembling and cloud-native open-source tools},
  author    = {Biderman, Dan and Whiteway, Matthew R and Hurwitz, Cole and
               Greenspan, Nicholas and Lee, Robert S and Vishnubhotla, Ankit and
               Warren, Richard and Pedraja, Federico and Noone, Dillon and
               Schartner, Michael M and others},
  journal   = {Nature Methods},
  volume    = {21},
  number    = {7},
  pages     = {1316--1328},
  year      = {2024},
  publisher = {Nature Publishing Group US New York}
}

Original data archive: https://doi.org/10.6084/m9.figshare.24993363.