Senior Computer Vision Engineer
Niantic SpatialBase Salary
$189k - $255k/yr
Responsibilities
- Design, develop, and maintain production-grade computer vision systems for the Visual Positioning System and 3D mapping pipelines.
- Refine and scale Structure from Motion, feature extraction, and descriptor extraction algorithms and move research concepts into production code.
- Optimize machine learning and computer vision code for low-latency, high-performance execution on cloud and mobile GPU/CPU environments.
- Create and own tools and frameworks for evaluating spatial grounding and 3D map quality against ground-truth data.
- Lead technical design reviews, mentor junior engineers, and resolve complex technical disagreements within the mapping stack.
- Collaborate with Product, Research, and Operations teams to deliver computer vision solutions and user experiences.
Requirements
- Bachelor’s, master’s, or PhD in Computer Science, Robotics, Computer Vision, or a related technical field, or equivalent professional experience.
- 5+ years of experience developing and shipping algorithms for image processing, computer vision, or 3D reconstruction.
- Expert-level proficiency in Python and/or C++.
- Demonstrated experience designing solutions for Structure from Motion, Visual Positioning Systems, or 3D mapping.
- Deep experience with PyTorch or JAX and Git.
- Preferred experience planning and leading technical projects from inception to production.
- Preferred contributions to open-source computer vision libraries such as OpenCV or COLMAP.
- Preferred experience with CUDA or shader programming for performance optimization.
Benefits
- Requires working in the office 3 days per week.
Categories
About Niantic Spatial
First, our founders brought digital mapping to the world. Next, we invented global-scale AR. Now we're building the real-world foundation model for people, AI and robots. We're building for the 80% of economic activity that takes place beyond our screens: robots that lose GPS and drift off course, defense teams with no shared 3D picture of where they operate, facilities where downtime costs millions. Unlike LLMs, which understand language, we build computer-vision models that understand physical space. And unlike generated environments, our models are geometrically accurate and have real-world coordinates - which is what makes physical AI precise and trustworthy. We work across three capabilities for our customers: reconstructing physical spaces as AI-native digital twins using cheap, off-the-shelf cameras; locating and orienting – “localizing” – machines and people within those spaces; and enabling AI to understand and answer questions about the real world.