CRIM13 — Lightning Pose Single-View Dataset

Single-camera pose estimation dataset for two-mouse interaction keypoints, packaged for use with Lightning Pose.

Dataset Description

The Caltech Resident-Intruder Mouse dataset (CRIM13; Burgos-Artizzu et al., CVPR 2012) consists of two mice interacting in an enclosed arena, captured by top and side view cameras at 30 Hz. Only the top view is used here. 14 keypoints are labeled — 7 body parts on each of two mice (black and white).

Each keypoint in the original CRIM13 dataset was labeled by five independent annotators. Final labels are the median across all annotators per keypoint. Frames where one or both mice were absent have been removed.

The InD/OOD split follows the original dataset's train/test split: the 4 resident mice appear in both sets, but the intruder mouse differs across sessions.

Source data: original archive at https://data.caltech.edu/records/4emt5-b0t10.

Data Splits

Split	Labeled frames	Sessions
In-distribution (InD)	3,986	37
Out-of-distribution (OOD)	1,274	19

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

Keypoints

14 keypoints total: 7 body parts × 2 mice.

Body part	Black mouse	White mouse
Nose	black_mouse_nose	white_mouse_nose
Right ear	black_mouse_right_ear	white_mouse_right_ear
Left ear	black_mouse_left_ear	white_mouse_left_ear
Top of neck	black_mouse_top_of_neck	white_mouse_top_of_neck
Right rear knee	black_mouse_right_rear_knee	white_mouse_right_rear_knee
Left rear knee	black_mouse_left_rear_knee	white_mouse_left_rear_knee
Base of tail	black_mouse_base_of_tail	white_mouse_base_of_tail

Directory Structure

crim13/
├── 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_crim13.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_crim13.yaml is a ready-to-use training config. Key settings:

• Image resize: 256 × 256
• Backbone: resnet50_animal_ap10k
• Keypoints: 14

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

litpose train config_crim13.yaml

Citations

If you use this dataset, please cite the original CRIM13 paper:

@inproceedings{burgos2012social,
  title        = {Social behavior recognition in continuous video},
  author       = {Burgos-Artizzu, Xavier P and Doll{\'a}r, Piotr and Lin, Dayu and
                  Anderson, David J and Perona, Pietro},
  booktitle    = {2012 IEEE Conference on Computer Vision and Pattern Recognition},
  pages        = {1322--1329},
  year         = {2012},
  organization = {IEEE}
}

and the MARS paper, whose authors collected the keypoint annotations:

@article{segalin2021mouse,
  title     = {The Mouse Action Recognition System (MARS) software pipeline for
               automated analysis of social behaviors in mice},
  author    = {Segalin, Cristina and Williams, Jalani and Karigo, Tomomi and Hui, May and
               Zelikowsky, Moriel and Sun, Jennifer J and Perona, Pietro and
               Anderson, David J and Kennedy, Ann},
  journal   = {eLife},
  volume    = {10},
  pages     = {e63720},
  year      = {2021},
  publisher = {eLife Sciences Publications, Ltd}
}

Original data archive: https://data.caltech.edu/records/4emt5-b0t10.