Computer Vision Engineer (Autonomous Driving)

The role

The Computer Vision Engineer researches and develops unified vision models and learning methods for autonomous driving. This role involves training models for 3D scene understanding tasks and optimizing distributed training across multiple GPUs and nodes.

What you will do

• Research and develop multi-task learning models and training strategies for autonomous driving perception tasks.
• Improve data composition, sampling, loss balancing, and training schedules.
• Identify and optimize computation, memory, communication, and data pipeline bottlenecks in large-scale training.
• Reproduce and analyze performance regressions and failure cases from real-world driving.
• Conduct ablation studies to distinguish effects of data, model architecture, and training conditions.
• Research and develop representation learning and self-supervised learning methods using unlabeled driving data.
• Design and validate pre-training recipes combining self-supervision, knowledge distillation, and auxiliary supervision.
• Investigate representation collapse and training instability.

Required qualifications

• Deep understanding of computer vision and deep learning
• At least 2 years of experience researching and developing models
• Understanding of 3D geometry, camera models, coordinate transformations, and multi-view geometry
• Experience applying 3D geometry and multi-view geometry to 3D vision model development
• Experience with multi-GPU distributed training and improving training performance, memory usage, and numerical stability
• In-depth research or development experience in at least one of multi-task learning, representation learning, or self-supervised learning

Nice to have

• Record of presenting research at top-tier conferences in computer vision, robotics, or machine learning
• Experience improving scaling efficiency and GPU resource utilization in large-scale multi-node training
• Experience addressing loss balancing, task interference, negative transfer, or other key challenges in multi-task learning
• Experience developing BEV- or PV-based 3D scene representations using multi-camera and temporal information

Before you apply

• A reference check may be conducted after the interview process with consent
• A 3-month probationary period may apply

About the company

42dot researches and develops unified vision models and learning methods for autonomous driving.