Efficiency
The whole pipeline is four steps — spend time labeling, not on project scaffolding and sampling boilerplate.
Basic Tools
Pose Labeler
Pose Dataset Annotation Software
Desktop pose labeling and training software for ethology — four clear steps from annotation to model training, with custom keypoints and in-house pose estimation.
4 stepsLabel to train
CustomKeypoints
SpeciesTemplates
WindowsSupported OS
CAPABILITIES
Lower the barrier to pose labeling
High-quality pose datasets are the textbook for AI. Existing tools are often too heavy — Pose Labeler folds project setup, frame sampling, labeling, and training into one clear chain.
Flexibility
Custom keypoint names, colors, and counts — or load mouse, monkey, dog, eye-tracking, and gait templates.
Accuracy
In-house pose estimation plus precise labels give tracking models a reliable data foundation.
WORKFLOW
Four steps from labels to a model
Pose labeling → model training and export → load in Analyzer and other BehaviorAtlas software.
01
Project & sampling
Create or open a project, import at least five representative videos (all views), and sample ~800–1000 frames by video length.
02
Keypoint setup
Configure keypoints manually or apply a template, then write into the project — optionally save as a reusable template.
03
Labeling
Zoom, click to place, drag to adjust; Ctrl+Z / Ctrl+X to undo or clear. Labels save automatically with progress on the sample table.
04
Train & export
Optional batch training and ResNet50/101; export the model and load it in Analyzer via config.yaml.
KEYPOINT TEMPLATES
Configurable across species and paradigms
Keypoint count follows the analysis need — solid objects in the scene can be labeled too.
Mouse · 16 pts
Default 16 points: nose, ears, neck, limb/claw, back, and tail root/mid/tip.
nose · left_ear · right_ear · neck · left_front_limb · right_front_limb · left_hind_limb · right_hind_limb · left_front_claw · right_front_claw · left_hind_claw · right_hind_claw · back · root_tail · mid_tail · tip_tail
Monkey · 21 pts
Includes eyes, shoulders, elbows, wrists, knees, and ankles for NHP pose.
nose · left_eye · right_eye · head · neck · left_shoulder · right_shoulder · left_elbow · right_elbow · left_wrist · right_wrist · left_limb · right_limb · left_knee · right_knee · left_ankle · right_ankle · back · root_tail · mid_tail · tip_tail
nose · left_ear · right_ear · neck · front_spine · mid_spine · back_spine · root_tail · top_tail · left_front_leg · left_front_paw · right_front_leg · right_front_paw · left_back_leg · left_back_paw · right_back_leg · right_back_paw
Mobile eye · 9 pts
Pupil center plus pupil1–8, labeled counterclockwise from the rightmost point.
pupil_center · pupil1 · pupil2 · pupil3 · pupil4 · pupil5 · pupil6 · pupil7 · pupil8
pupil1–8 labeled counterclockwise from the rightmost pupil point
Head-fixed eye · 11 pts
The 9-point pupil set plus left/right canthus.
pupil_center · pupil1 · pupil2 · pupil3 · pupil4 · pupil5 · pupil6 · pupil7 · pupil8 · canthus_left · canthus_right
pupil1–8 labeled counterclockwise from the rightmost pupil point
Gait · 17 pts
Shoulder/elbow/wrist/toetip and hip/knee/ankle/toetip for motor analysis.
back · left_front_shoulder · left_front_elbow · left_front_wrist · left_front_toetip · right_front_shoulder · right_front_elbow · right_front_wrist · right_front_toetip · left_hind_hip · left_hind_knee · left_hind_ankle · left_hind_toetip · right_hind_hip · right_hind_knee · right_hind_ankle · right_hind_toetip
SPECIFICATIONS
Light labeling, GPU when you train
Labeling runs on a regular office PC; GPU is recommended for training.
Labeling hardwarei5-9400 · 4 GB RAM · office laptop OK
Training (recommended)i9-11900K · 32 GB · RTX 3090/4090
OSWindows 10 / 11
BackboneMouse ResNet50 · monkey/dog ResNet101
Frame budgetAbout 800–1000 frames per project
Model useExport, then load config.yaml in Analyzer
FAQ
Label quality drives tracking
Label quality directly drives tracking quality.
Yes — any solid-volume object is worth tracking.
Usually too few, inaccurate, or unrepresentative labels. Add samples/frames, relabel, and retrain.
PART-TIME
Join as a part-time pose labeler
We hire part-time data labelers on an ongoing basis. Below is a typical task brief — we confirm details and arrange training after you apply.
Get trained
We send the project link and train you on the tool and keypoint conventions first.
Sample review
Submit 2–3 labeled samples for review; continue in batch only after approval.
Batch labeling
Label carefully and accurately; follow keypoint order, with left/right from the mouse’s own sides.
How to apply
Apply via the Contact form and mention “part-time labeling” in your message. We will follow up and screen applicants.
Request a Pose Labeler software solution
Share your species, keypoints, and labeling scenarios — get guidance on software setup, model training, and deployment.