Sr. Machine Learning Engineer (Perception and Tracking)

At Ouster, we build sensors and tools for engineers, roboticists, and researchers, so they can make the world safer and more efficient. We've transformed LIDAR from an analog device with thousands of components to an elegant digital device powered by one chip-scale laser array and one CMOS sensor. The result is a full range of high-resolution LIDAR sensors that deliver superior imaging at a dramatically lower price. Our advanced sensor hardware and vision algorithms are used in autonomous cars, robotics, industrial, and smart infrastructure applications (among many others). If you’re motivated by solving big problems, we’re hiring key roles across the company and need your help!
We are looking for a highly technical Machine Learning Engineer to lead our efforts in Object Detection and Tracking. You will not simply be "importing" pre-made models; you will be architecting deep neural networks, translating state-of-the-art research papers into code, and optimizing these systems for real-time, on-device performance.

This role requires a deep knowledge of neural network architectures. You should be confident ripping apart a model to modify layers, loss functions, and data flows to fit our specific constraints.

• Architect Unified Models: Design and train DNN models that perform Object Detection and Tracking simultaneously, leveraging temporal information to improve consistency.
• Research to Production: Evaluate state-of-the-art research papers and prototype these concepts (turning papers into code) and adapt them into robust, production-grade solutions.
• Deep Model Customization: Go beyond standard libraries by implementing custom loss functions, modifying internal model architectures, and designing specific data augmentation strategies to squeeze out maximum performance.
• Edge Optimization: Ensure high accuracy is matched by high efficiency. Optimize models for real-time inference and on-device deployment.
• Data Strategy: Develop training recipes for data-constrained environments and effective post-training strategies.

• Core Stack: 
  • 5+ years proficiency in Python and PyTorch. 
  • 3+ years proficiency in C++ for production deployment and optimization.
• Detection & Tracking: Deep theoretical and practical understanding of modern object detectors (e.g., Transformers, YOLO variants, R-CNNs) and tracking algorithms (e.g., DeepSORT, Kalman Filters, Optical Flow).
• Architecture Internals: Proven experience not being dependent on "out-of-the-box" APIs. You have a track record of modifying model architectures via extensive experimentation to meet specific requirements.
• Low-Data Regimes: Experience improving model generalization with limited data using Transfer Learning, Domain Adaptation, or Few-Shot Learning.
• Mathematical Foundation: Strong grasp of linear algebra and probability as it applies to custom loss function design and geometric 3D vision.

• 3D / LiDAR Experience: Hands-on experience with 3D Point Cloud data (LiDAR) is a massive plus.
• Deployment Tools: Experience with TensorRT, ONNX Runtime, or edge-specific hardware (NVIDIA Jetson, etc.).