9 hours ago
Responsibilities
- Build perception pipelines for depth, pose, tracking, segmentation, and 3D reconstruction over large video and multi-sensor datasets.
- Apply SLAM, visual odometry, and structure-from-motion to recover camera trajectories and scene geometry at scale.
- Train, fine-tune, or deploy vision and vision-language models for embodied tasks such as hand-object interaction and scene understanding.
- Develop automated QA and filtering systems for calibration errors, synchronization drift, motion blur, and other data defects.
- Create benchmarks and evaluation frameworks for measuring dataset impact on robotics and world-model performance.
- Work with frontier labs and robotics teams to translate data requirements into production pipelines and own problems through deployment.
Requirements
- At least 3 years of production perception or computer vision experience for robotics, AR/VR devices, autonomous vehicles, or similar applications.
- Hands-on experience with SLAM, visual-inertial odometry, multi-view geometry, or 3D reconstruction.
- Strong Python skills and fluency with PyTorch; C++ is a plus.
- Experience with multi-sensor data, including camera intrinsics and extrinsics, IMU, depth, and time synchronization.
- Experience training, fine-tuning, or deploying modern deep learning models for vision.
- Strong evaluation and edge-case intuition, with the ability to own work from customer discussions through deployed pipelines.
- Bonus qualifications include neural scene representation experience such as NeRFs or Gaussian splatting, egocentric or teleoperation data experience, and publications or open-source work in perception, 3D vision, or robotics.
Benefits
- Onsite at the San Francisco headquarters five days per week.
Categories
ML EngineeringRobotics
About Sieve
Sieve builds datasets and infrastructure for training video AI models, combining large-scale video processing with novel video understanding techniques. Its tools support ML teams working on media, gaming, AR/VR, and robotics by supplying high-quality video training data and pipelines. Founded in 2016 and headquartered in Palo Alto, the company operates as a privately held AI lab focused exclusively on video data.