Per-keypoint coordinates
X/Y position and confidence score for each keypoint — head, body center, tail base, and four paws — at every frame. Enables post-hoc computation of any body-part-derived measure.
CSVJSON
← Capabilities
Pose Estimation
Keypoint-level tracking of body parts across species and paradigms
Go beyond center-of-mass — track head direction, paw placement, tail position, and body orientation frame by frame.
7+
Keypoints tracked
Multi
Species supported
Frame-byframe
Temporal resolution
SuperAnimal
Pretrained model library
The problem
Center-of-mass tracking discards body-part-level information
A single centroid coordinate cannot represent the complexity of rodent posture and movement. Head direction is lost, making it impossible to determine what the animal is investigating. Paw placement is invisible, preventing gait quantification from standard video. Postural changes — hunching, stretching, twisting — are collapsed into a single point.
- Head direction is unknown — approach toward a stimulus cannot be distinguished from backing away
- Paw placement and limb extension are invisible to centroid tracking, precluding gait-level analysis
- Postural state — hunched, stretched, curled — carries behavioral meaning that a single coordinate discards
The solution
Multi-keypoint pose estimation with pretrained and custom models
ConductVision tracks 7+ body keypoints — head, body center, tail base, and four paws — using SuperAnimal pretrained models. Custom keypoint configurations can be trained for specialized preparations. All keypoint coordinates are exported per frame for downstream analysis.
- Seven or more keypoints tracked simultaneously — head, body, tail, and all four paws
- SuperAnimal pretrained models provide immediate deployment across standard rodent species
- Custom models trainable for non-standard species or specialized keypoint configurations
Endpoints
Body-part-level dependent variables
Body orientation and limb extension metrics
Head-body angle, body axis orientation, limb extension, and tail curvature computed from keypoint geometry. Pre-computed for common analyses.
CSV
Posture classification per frame
Discrete posture labels — rearing, grooming, hunched, stretched attend — assigned per frame from keypoint configurations.
CSV
Applications
Paradigms requiring body-part-level resolution
Head direction in maze choice points
Track where the animal is looking, not just where its body center is located. Head direction at T-maze or Y-maze choice points reveals vicarious trial-and-error and decision dynamics.
Measures
- Head direction at choice point
- Head scan frequency
- Gaze-movement coupling
Paw placement in gait studies
Track individual paw positions frame by frame for stride length, stance width, and step sequence analysis from standard overhead or side video.
Measures
- Stride length per paw
- Stance width
- Step sequence regularity
Postural assessment in neurological models
Quantify postural abnormalities — asymmetric limb extension, tail deflection, body axis curvature — in stroke, TBI, or neurodegenerative models.
Measures
- Limb extension asymmetry
- Tail deflection angle
- Body axis curvature
Body axis orientation and turning behavior
Compute heading direction, turning rate, and angular velocity from body axis orientation. Detect ipsiversive and contraversive turning in unilateral lesion models.
Measures
- Heading direction
- Turn rate
- Ipsiversive/contraversive ratio
Compared to typical systems
How ConductVision differs
| Feature | ConductVision | Typical systems |
|---|---|---|
| Tracking resolution | 7+ keypoints per animal | Single centroid |
| Species coverage | Mouse, rat, and custom species | Species-specific licenses |
| Pretrained model availability | SuperAnimal models included | Requires manual annotation and training |
| Custom keypoint training | Supported — GUI-based annotation | Limited or not available |
| Head direction output | Computed from head-body axis | Not available from centroid |
Related capabilities
Gait Analysis
HMM-based stride detection from standard video with per-paw metrics — no treadmill or pressure plate required.
Behavior Classification
ML-powered frame-level behavioral classification with temporal smoothing and custom classifier training.
Precision Tracking
High-resolution 30 fps tracking that captures sub-second behavioral events conventional systems miss.
Track what center-of-mass tracking misses
See head direction, paw placement, and postural dynamics extracted from your existing video.
