Skip to content

Sample Datasets

Labeled datasets to practice on, compare against, or start a project from.

The tutorial videos

The Tutorial uses two videos of a pair of mice in a home cage, from the sleap-tutorial-data repository:

File Frames Used for
mice.mp4 14,100 (5 min at 47 fps) Labeling and training. Use sample video in New Project downloads it for you
new_data/mice_new.mp4 2,820 (1 min) Running a trained model on a new video

The SLEAP paper datasets

These are the labeled datasets from the SLEAP paper (Pereira et al., 2022). Each is a .pkg.slp labels package, with the images embedded, so it opens in the app on its own (File ▸ Open Project…). Each comes split into training, validation and test sets.

Dataset Animal Nodes Animals per frame Identity Labeled instances
fly32 Fruit fly, single, animal-centered 32 1 no 1,500
flies13 Fruit flies, courting pairs 13 2 yes 4,000
mice_of Mice in an open field, filmed from below 11 1–5 no 2,950
mice_hc Mice in a home cage, filmed from above 5 2 no 2,948
bees Bumblebees in a petri dish 21 2 no 1,604
gerbils Gerbils, pups and adults, in a home cage 14 4 yes 1,588

"Identity" means the animals can be told apart by appearance, so these datasets suit the + ID model types.

The animal skeleton templates in New Project and the Skeleton panel are these datasets' skeletons, so you can start a similar project with the same body parts.

Download links, frame rates, resolutions and credits for each dataset are on the SLEAP docs' Datasets page. The full labeled data and trained models from the paper are on OSF.