Applied Computer Vision Engineer

Holstein Association USA, the world’s largest dairy breed organization, is seeking a highly motivated and innovative Applied Computer Vision Engineer to contribute to the advancement of automated phenotyping of dairy cattle using computer vision. This unique opportunity allows the right candidate to join a transformative project that leverages cutting-edge deep learning techniques to revolutionize how conformation traits are evaluated in Holstein cattle-both phenotypically and genetically.

The selected candidate will work with an established and functional computer vision pipeline and be responsible for leading the refinement, validation, and optimization of trait estimation algorithms.

This is a limited-term position funded for 3 years, with the potential for renewal or continuation depending on performance and availability of funding.

Holstein Association USA records linear type traits on Holstein cows across the United States through a program of trained classifiers. This project expands that scoring toward image-based estimation, working from a camera and depth-sensor pipeline that already runs in the field. You would own the modeling and geometry work that turns raw captures into trait estimates. The traits you estimate feed the conformation evaluations breeders use to make mating and culling decisions, so accuracy and repeatability carry real weight. This is a lead role with room to set the technical direction.

The work centers on computer vision and machine learning. You would build, train, and validate deep learning models in Python and PyTorch that read 3D images, video, and point clouds of dairy cattle. Day to day, the role spans model training and evaluation, semantic and instance segmentation, object detection, keypoint and pose estimation, 3D reconstruction, camera and depth-sensor calibration, and model deployment. It sits at the meeting point of applied artificial intelligence, precision livestock farming, and dairy cattle genetics.

Key Responsibilities:

· Model development & trait estimation

o Lead the development and refinement of deep learning models for estimating dairy cattle linear conformation traits (e.g., stature, udder depth) from 3D image and video data

o Train, fine-tune, and benchmark landmark/key point detection and segmentation models (e.g., DeepLabCut-style pose estimation, CNN/transformer architectures) against manually scored ground truth

o Own the model training lifecycle: dataset curation, augmentation, hyperparameter tuning, cross-validation, and experiment tracking for reproducibility

· 3D / point-cloud processing

o Maintain and improve the 3D geometry pipeline: depth-to-XYZ reconstruction, coordinate-frame leveling and fiducial-based origin alignment, point-cloud filtering and registration (e.g., PCL / Open3D), and calibration/scale validation

o Diagnose and correct systematic measurement errors (e.g., scale, floor-tilt, and offset artifacts) that propagate into trait estimates

· Validation & quality assurance

o Validate model outputs against manually classified phenotypes and existing genetic evaluations; quantify accuracy, bias, and repeatability against trained-classifier ground truth.

o Support data preprocessing, annotation standardization, and large-scale QA of image/point-cloud datasets.

· Deployment & tooling

o Package and deploy trained models for batch and field use, with attention to GPU/CUDA performance, runtime efficiency, and reproducible inference (e.g., OpenCV, ONNX, containerized deployment)

· M.S., Ph.D. or equivalent experience in Computer Vision, Animal Science, Agricultural Engineering, Computer Science, or a closely related discipline

· Proven expertise in deep learning frameworks (PyTorch, TensorFlow) and model development in Python, including hands-on model training and evaluation (not only inference)

· Demonstrated experience with image-based phenotyping, computer vision applications, or related fields

· Demonstrated publication record or project leadership in relevant scientific or technical areas

Preferred Qualifications:

· Hands-on experience with 3D data and point clouds, depth sensors, point-cloud libraries (PCL, Open3D),registration, calibration, and coordinate-frame/geometry handling.

· Experience with pose/keypoint estimation for animals (e.g., DeepLabCut) or comparable landmark-detection methods

· Experience working with livestock imagery or behavior analysis in precision livestock farming

· Understanding of the dairy industry and linear classification systems for dairy cattle

· Familiarity with sensor fusion and large-scale dataset management.

· Experience with collaborative software development and cloud computing environments (e.g., AWS), containerization (Docker), and parallel/distributed processing

Skills and tools: Python, PyTorch, TensorFlow, OpenCV, deep learning, machine learning, neural networks, CNN and transformer architectures, semantic and instance segmentation, object detection, keypoint and pose estimation, 3D reconstruction, point-cloud processing (PCL, Open3D), depth sensors and LiDAR, camera calibration, sensor fusion, MLOps and experiment tracking, model deployment (ONNX, Docker), GPU and CUDA, AWS, and large-scale data annotation and quality assurance.

$90,000 - $110,000 a year - Full-time, Pulled from the full job description

• Referral program
• Tuition reimbursement
• 401(k)
• Health insurance
• 401(k) matching
• Paid time off
• Dental insurance, · A chance to directly impact the future of dairy cattle conformation evaluation with state-of-the-art tools

· Access to decades of pedigree and phenotypic data, extensive field infrastructure, and collaborative partners in industry and academia

· The opportunity to work on a pioneering phenotyping system already in place, with the mandate to lead the next stages of refinement and deployment

· Competitive salary and benefits package, commensurate with experience and tailored to a fixed-term research role

· Support for conference travel and peer-reviewed publication of your results

· Ground-truth scores from trained classifiers, plus decades of recorded phenotypes and pedigrees, to train and validate against

· A clearly scoped three-year mandate with ownership of the modeling and geometry work, and a direct line to the people who use the results, * 401(k)

• 401(k) matching
• Dental insurance
• Disability insurance
• Employee assistance program
• Flexible spending account
• Health insurance
• Life insurance
• Paid time off
• Referral program
• Tuition reimbursement