aquapose 1.0.0

3D fish pose estimation via differentiable refractive rendering

AquaPose

3D fish pose estimation via differentiable refractive rendering. AquaPose fits a parametric fish mesh to multi-view silhouettes from a 13-camera aquarium rig, producing dense 3D trajectories and midline kinematics for behavioral research on cichlids.

Installation

pip install aquapose

Quick Start

from aquapose.calibration import load_calibration
from aquapose.segmentation import segment_frame
from aquapose.optimization import optimize_pose

# Load multi-camera calibration (from AquaCal)
cameras = load_calibration("calibration.json")

# Segment fish in a multi-view frame
masks = segment_frame(frame, cameras)

# Reconstruct 3D pose via analysis-by-synthesis
pose = optimize_pose(masks, cameras)

Development

# Set up the development environment
pip install hatch
hatch env create
hatch run pre-commit install
hatch run pre-commit install --hook-type pre-push

# Run tests, lint, and type check
hatch run test
hatch run lint
hatch run typecheck

See Contributing for full development guidelines.

Documentation

License

MIT