Background:
Xihelm is developing state-of-the-art robotics for handling fruit and vegetables. You’ll be helping to make that a reality - building high quality vision systems to detect bad fruit before they get packed and hit the shelves - we've got real challenges to solve. It's a great opportunity to learn more about cutting edge robotics, machine learning and computer vision.
Responsibilities:
- Define, build and independently deliver a visual-based quality-assurance system for fruit with >99% accuracy at <1sec processing using a combination of multispectral imagers, handling difficult/partial viewpoints, hard-to-see damage and fragile crop - improving it over a multi-year, funded project lifecycle, and communicating well throughout.
Notes:
At Xihelm we view that most of the time, off the shelf and published techniques can work quickly, and we want people who embrace (but are willing to invalidate) standard techniques first - who understand delivery risk given uncertainty around training set sizes, etc.
All successful candidates are likely to be asked to do an at-home challenge of 2-3hrs - this is a fair opportunity to show you at your best.
This is a non-negotiable onsite role in London UK, though we can sponsor visas.
Note: No recruiters (we aren't responsible for any fees/invoices)
Requirements
Must have experience:
- Developed computer vision algorithms for practical and difficult feature detection in commercial settings as an independent researcher.
- Knowledge of state of the art algorithms & relevant research - and a passion to apply them and stay on top of the SOTA
- Be able to show practical knowledge of data augmentation, hyperparameter tuning, GANs and other relevant techniques
- Be shown to be able to quickly create & curate relevant datasets from a standing start
- Great quality demonstrable experience with classical computer vision, machine learning (supervised, unsupervised) and modern deep neural networks (including transfer learning) for image-related segmentation, classification etc.
- Ability to write good Python code using e.g. Tensorflow, Keras and/or pyTorch.
- A good compass on the “right way to go”, documenting as you go, with the ability to abandon paths of research (given diminishing returns etc) and handle critical feedback from colleagues well.
- The ability to work with minimal supervision, but whilst communicating well (verbally & in writing) to the team and beyond, and accepting critical feedback well.
Preferred experience:
- Performed multi-camera registration and calibration in the real world (including lighting).
- Experience with hyperspectral, multispectral imagers (VNIR etc), and depth cameras, along with practical integration of vision-based systems (on e.g. conveyor belts) for quality assurance or damage detection
- Experience with 3D via RGBd, IRd, and stereo
- 3D reconstruction experience (especially from multiple cameras and/or angles)
- ROS experience (your work will likely end up on our robots)
- Experience with labeling tools, and custom UIs to collect training sets
- Comfortable in Bash, Git, Linux, Docker, Jenkins
- Can read C and C++, and handle dependency hell
- Exposure and comfort in Continuous Integration and test-driven development
- Cloud computing (e.g. model training on GPUs)
- Defined labeling tasks for external labellers
- Writing reports for sharing with external stakeholders (e.g. research funding agencies).
- Advanced troubleshooting skills
- Broader experience with sensor fusion, LIDAR and SLAM
- Rig build skills (not afraid to make a quick camera mount)
- Publication track record at conferences e.g. CVPR, ICCV, ECCV etc.
- Completed advanced level studies (MSc or PhD) in Computer Vision, Robotics, Computer Science, Electrical Engineering, mathematics or Physics or a related technical field.
Benefits
- Salary: £80-140k FTE, +stock options etc.
- Playing with robots
- Medical, dental and life insurance
- Frequent free lunches, and a well stocked snack kitchen
- 25 working days leave annually
- Tuition reimbursement
