Job description
About the role
The Innomium Vision program publishes compact detection artifacts and applies the same discipline to real operating environments. We treat camera conditions, data quality, runtime constraints, and workflow behavior as part of the model problem.
The mandate
You will own computer-vision work from data and baseline design through training, error analysis, model export, runtime profiling, and pilot evidence. Projects may involve object detection, segmentation, event logic, edge deployment, or adaptation of public releases such as Sentinel, Vantage, and Ember.
This role is not only about improving a headline score. You will investigate which scenes fail, how object scale and occlusion change outcomes, whether post-processing helps, and what an alert should mean to the operator.
What strong performance looks like
You create evaluation protocols that expose the difficult tail, produce reproducible training and inference artifacts, and explain the accuracy-latency-size trade-off clearly. You can move a promising model into a bounded pilot without overstating what the evidence proves.
How we work
You will partner with research, data, product, and infrastructure engineers. We expect careful experiment records, pragmatic model choices, and respect for privacy, safety, and the limits of computer vision in high-consequence workflows.
Responsibilities
The work this role is expected to own.
- Design representative datasets, splits, annotation guidance, and scene-level evaluation protocols
- Train, adapt, distill, and compare detection or segmentation models for target environments
- Perform structured error analysis across camera, condition, class, scale, and failure mode
- Export and validate ONNX or other deployment artifacts and profile complete inference pipelines
- Collaborate on temporal logic, event semantics, review interfaces, and monitoring
- Document model lineage, data limitations, runtime assumptions, and production acceptance evidence
Requirements
Capabilities and experience that support success in this role.
- Professional experience developing and evaluating modern computer-vision systems
- Strong Python, PyTorch, data-pipeline, and experiment-management skills
- Practical understanding of detection metrics, dataset bias, augmentation, and error analysis
- Experience taking models into a runtime outside the training environment
- Ability to reason about latency, memory, hardware, privacy, and operational consequences
- Evidence of reproducible technical work through code, models, papers, demos, or shipped systems
Nice to have
Useful adjacent experience, but not a substitute for the core requirements.
- Experience with YOLO-family models, ONNX Runtime, TensorRT, OpenVINO, or edge accelerators
- Experience with tracking, camera geometry, browser inference, or video pipelines
- Background in industrial, logistics, safety, or embedded-vision environments
How to apply
Send a concise introduction connecting your experience to the mandate. Include links to shipped, published, measured, or inspectable work, and identify the decisions or tradeoffs you personally owned.
Compensation, engagement structure, benefits, jurisdiction, eligibility, and working-time overlap are discussed early in the process. Generic cover letters are not required.
Email your application