Behavior & Pose Analytics

IntegraPose

A unified desktop application for pose estimation, behavior classification, and downstream analytics - built for real lab workflows.

Train or import a model, run inference on new recordings, measure behavior and ROI events, check predictions against the video, and explore sub-behaviors inside known classes - without stitching together a collection of separate tools.

Open Quick Start See Pose Workflow See Detection-Only Workflow

Detection Pose Behavior classification Batch analytics ROI metrics Manual review Sub-behavior discovery Custom architectures Plugin-ready

What the app covers

• Project setup and annotation
• Pose-model training and model import
• File and webcam inference
• Behavior, ROI, and object-bout review
• Sub-behavior discovery
• Custom YOLO architectures

Where IntegraPose Fits

Computational ethology has matured into a rich ecosystem of specialized tools. Pose estimation has strong open-source options such as DeepLabCut and SLEAP. Unsupervised behavior discovery has B-SOiD, VAME, and Keypoint-MoSeq. Manual event coding is well served by BORIS, and commercial suites cover regulated end-to-end work. Each is excellent at what it does - and many labs still have to assemble several tools, with custom scripts in between, to move from raw video to behavior measures ready for interpretation.

IntegraPose addresses the seams in that workflow. It brings pose estimation, multi-animal tracking, ROI- and bout-level analysis, video-guided manual review, and optional sub-behavior discovery into one desktop application. The aim is not to replace every specialized tool. It is to give labs without dedicated engineering support a clear, reproducible path from raw video to results they can inspect and defend.

What You Can Build with IntegraPose

The same workflow pattern adapts across different research and movement- analysis settings.

Gait & Kinematic Analysis

Quantify stride length, speed, paw angle, and other locomotion features to study movement in health and disease.

Real-time Behavior Apps

Drive closed-loop experiments, biofeedback, and live monitoring with low-latency pose + behavior streams.

Rodent Assay Workflows

Score bouts, ROI occupancy, and inter-animal interactions across standard rodent paradigms.

Sports & Movement Analytics

Apply the same pose + behavior pipeline to athletic performance, technique review, or rehabilitation.

Browse more example outputs

Start With The Right Path

If you want to...	Start here	Best fit
Learn the layout and run a first project	Quick Start	New users
Use an existing detection model and skip pose training	Detection-Only Model Workflow	Detection-first workflows
Train and use a pose model inside IntegraPose	Pose Model Workflow	Full pose workflows
Process many recordings at once	Batch Processing Wizard	High-throughput labs
Check and correct predicted behavior or ROI events	Bout Review Workspace	Studies that include manual review
Find and interpret batch result files	Batch Output Map	Completed batch runs
Design a custom YOLO architecture for your assay	Customizing the YOLO Model	Power users
Explore optional tools and plugins	Plugin Catalog	Extended workflows

Workflow At A Glance

Stage	Main result
Data Preprocessing	Extracted frames, cropped videos, organized source folders
Setup and Annotation	Project settings, classes or keypoints, dataset.yaml
Model Training	Trained YOLO pose-model files and training measurements
Inference	Detection or pose labels, videos, optional motion summaries
Bout & ROI Analytics	Behavior bouts, ROI measures, object interactions, and review-ready results
Batch Processing Wizard	Repeated analytics runs across many videos
Behavior Clustering	Candidate movement patterns within model classes, review scores, and optional named bout clips (pose workflows)

Raw videos
  -> Data Preprocessing
  -> Setup and Annotation
  -> Model Training (or imported model, or custom architecture)
  -> Inference or Batch Processing Wizard
  -> Bout & ROI Analytics
  -> Manual review (when required by the study)
  -> Behavior Clustering (optional)

Going Further

When the standard tabs are not quite enough:

• Customize the YOLO architecture - edit the model .yaml to swap backbones, fuse modules differently, add attention or transformer blocks, or tune for edge deployment. CLI training instructions included.
• Behavior Clustering - explore candidate movement patterns within a YOLO class, inspect them against the video, and export reviewed names and bout clips for further analysis.

The Plugin Ecosystem

IntegraPose includes a curated plugin ecosystem for needs outside the seven main tabs. Each plugin is optional: enable the ones you need from Plugins -> Manage Plugins..., launch them from the Plugins menu, and keep the core workflow focused on your experiment.

Plugin status - research in progress

The plugin ecosystem evolves with active research. Some plugins are stable, others are works in progress, and the set may change as research priorities shift. See the Plugin Catalog for the current status note and per-plugin guides.

Dataset creation

Assisted Pose Curation - review-first pose labeling with model-assisted suggestions.
AutoLabel Forge - GroundingDINO + SAM-assisted auto-labeling for detection datasets.
Dataset Augmentor Lab - GUI-driven augmentation for YOLO datasets.

Behavior & sequence modeling

TandemYTC - Tandem YOLO + Temporal Classifier - full-video annotation, YOLO-pose review overlays, temporal-model training, and bounded-latency inference.

Domain-specific analytics

Gait & Kinematic Dashboard - stride length, speed, paw angle, and locomotion comparisons.
Fura Imaging Lab - Fura-2 stack alignment, ROI tracking, ratio analysis, and workbook export.
Zone Counter - live polygon-based zone counts during inference.

Exploration & review

EDA Tool - interactive PCA / KMeans on pose embeddings with video sync.

See the full Plugin Catalog →

Compatibility Notes

• Inference supports both detect and pose file-based workflows.
• Model Training is pose-oriented in the GUI. Detection checkpoints are imported into Inference; detection-model training is outside the built-in training tab.
• Bout Analytics works with both detection-only and pose label outputs.
• Bout Review Workspace reviews Class ID behavior bouts, concurrent ROI, exclusive ROI-X, and pose-based object interactions from completed analytics runs.
• Behavior Clustering (Tab 7) is pose-only; it accepts pose data, Bout Analytics output, or batch manifests as input.
• Batch Processing Wizard is available from File -> Batch Processing Wizard....
• Optional plugins can be enabled from Plugins -> Manage Plugins....

Open Source Foundations

IntegraPose builds on open-source projects that make modern vision and analytics workflows practical for research labs.

Project	Role in IntegraPose
PyTorch	Deep-learning runtime used by model workflows and GPU-backed inference stacks
Ultralytics YOLO	Core training and inference backbone for pose and detection workflows
OpenCV	Video IO, image processing, overlays, and supporting CV utilities
NumPy and SciPy	Numerical processing across training, analytics, and feature computation
Pandas	Tables, bout summaries, and export-friendly data handling
Matplotlib	Plotting and reporting visuals
Pillow	Image loading, export, and GUI-friendly image utilities
Supervision	Overlay and workflow helpers for modern computer-vision pipelines

Read citations and acknowledgements