sleap-io 0.9.0

Standalone utilities for working with pose data from SLEAP and other tools.

sleap-io

A standalone Python library and CLI for working with animal pose tracking data. Read, write, convert, and manipulate pose data across formats with minimal dependencies.

Complements the core SLEAP package but does not include labeling, training, or inference.

Documentation | Examples | CLI Reference

Features

• Multi-format I/O -- Read and write SLEAP, NWB, COCO, DeepLabCut, Ultralytics YOLO, JABS, Label Studio, CSV, Analysis HDF5, AlphaTracker, and LEAP formats
• CLI tools -- Inspect, convert, render, and transform data from the command line (sio show, sio convert, sio render, sio transform)
• Rendering -- Produce publication-quality videos and images with pose overlays, customizable colors, markers, and presets
• Transforms -- Crop, scale, rotate, pad, and flip videos with automatic coordinate adjustment
• Merging -- Combine annotations from multiple sources with flexible matching strategies
• Codecs -- Convert to/from NumPy arrays, DataFrames (pandas/polars), and dictionaries
• Video I/O -- Read any video format via pluggable backends (FFMPEG, OpenCV, PyAV) with a NumPy-like interface
• Lazy loading -- Load large SLP files up to 90x faster by deferring object creation
• Remote URLs -- Load .slp files directly from https://, s3://, gs://, and az:// URLs with lazy range-based reads and optional persistent caching
• Dataset splits -- Create train/val/test splits and export to formats like Ultralytics YOLO

Installation

Quick start (no install needed)

Run CLI commands instantly with uvx:

uvx sleap-io show labels.slp
uvx sleap-io convert -i labels.slp -o labels.nwb

Install as CLI tool

uv tool install "sleap-io[all]"
sio show labels.slp

Install as Python library

pip install "sleap-io[all]"
# or: uv add "sleap-io[all]"
# or: conda install -c conda-forge sleap-io

From source

pip install "sleap-io[all] @ git+https://github.com/talmolab/sleap-io.git@main"

Optional extras

Video support works out of the box via imageio-ffmpeg. Optional extras provide faster backends and additional format support:

Extra	Purpose
opencv	Faster video backend
pyav	Alternative video backend
mat	LEAP .mat file support
polars	Fast DataFrame operations
all	All of the above

Development

git clone https://github.com/talmolab/sleap-io.git && cd sleap-io
uv sync --all-extras

See CONTRIBUTING.md for more details.

Usage

CLI

# Inspect a labels file
sio show labels.slp

# Convert between formats
sio convert -i labels.slp -o labels.nwb

# Render video with pose overlays
sio render -i predictions.slp -o output.mp4

# Transform (scale, crop, rotate) with coordinate adjustment
sio transform labels.slp --scale 0.5 -o scaled.slp

Python

Load and convert between formats

import sleap_io as sio

labels = sio.load_file("predictions.slp")
labels.save("predictions.nwb")

Format is auto-detected from the extension. See supported formats.

Convert to NumPy arrays

labels = sio.load_file("predictions.slp")

trx = labels.numpy()  # (n_frames, n_tracks, n_nodes, 2)
trx_with_scores = labels.numpy(return_confidence=True)  # (n_frames, n_tracks, n_nodes, 3)

Create labels from scratch

import sleap_io as sio
import numpy as np

skeleton = sio.Skeleton(
    nodes=["head", "thorax", "abdomen"],
    edges=[("head", "thorax"), ("thorax", "abdomen")]
)

instance = sio.Instance.from_numpy(
    points=np.array([[10.2, 20.4], [5.8, 15.1], [0.3, 10.6]]),
    skeleton=skeleton
)

video = sio.load_video("test.mp4")
lf = sio.LabeledFrame(video=video, frame_idx=0, instances=[instance])
labels = sio.Labels(videos=[video], skeletons=[skeleton], labeled_frames=[lf])
labels.save("labels.slp")

Render poses

labels = sio.load_file("predictions.slp")
labels.render("output.mp4")                      # Full video
labels.render("preview.mp4", preset="preview")   # Fast 0.25x preview
sio.render_image(labels[0], "frame.png")         # Single frame

Merge annotations

base = sio.load_file("manual_annotations.slp")
predictions = sio.load_file("predictions.slp")
base.merge(predictions)
base.save("merged.slp")

Create training splits

labels = sio.load_file("labels.slp")
labels.make_training_splits(n_train=0.8, n_val=0.1, n_test=0.1, save_dir="splits/", seed=42)

See the Examples page for more recipes including NWB export, video re-encoding, skeleton replacement, path fixing, and YOLO/COCO export.

Support

For technical inquiries, please open an Issue.

For general SLEAP usage, see sleap.ai.

License

BSD 3-Clause License. See LICENSE for details.