公司介绍

Bosch China Innovation and Software Development Campus

博世创新与软件开发中心

    博世创新软件开发（无锡）有限公司针对交通出行的电动化、自动化、互联化、个性化，提供面向未来的创新技术和前沿解决方案，加速针对中国市场的技术战略的实现和发展。博世软件中心主要发展方向包括智能网联汽车、智能座舱、自动驾驶、车路云协同、工业4.0、人工智能大数据、智能家居、嵌入式软件服务。博世中国创新与软件开发中心以软件为客户赋能，在汽车自动驾驶、氢燃料电池、重卡电驱动桥、多合一的控制器等多个领域取得创新研发成果。

职位描述Develop, design, and optimize deep learning networks for 3D/4D object /OCC recognitionConduct research on key issues in multi-task learning, including loss balancing and task conflict resolution, to improve network stability and generalization capabilityResearch and develop tracking algorithms for dynamic objects (e.g., pedestrians, vehicles, animals) and continuously optimize their performanceCollaborate with global team members to deploy and optimize deep learning networks on defined SoCs; possess in-depth understanding of post-training quantization, quantization-aware training (QAT), and pruning techniquesPerform data mining, scene classification, and preprocessing for autonomous driving object detection tasks; participate in dataset construction, annotation, and training sample management to enhance algorithm performance.

职位要求Master’s degree or above in Computer Science, Automation, or related fields>5 years working experience in computer vision for meachine learning in low computation power SOCKnowledgeable in common perception algorithms (e.g., Transformer, BEV, OCC, etc.Proficient in Python and C++ with strong programming skills; familiar with common machine learning/deep learning frameworks such as TensorFlow and PyTorchExperience in monocular vision projects is preferred, such as 3D perception based on monocular images, SLAM, or temporal modelingFamiliar with model optimization and acceleration techniques; experience in model deployment is a plusUnderstanding of model quantization(PTQ,QAT), capability of tuning the model structure to fulfil the specific SoC runtime requirementUnderstanding of multi-task learning principles and related techniques (e.g., parameter sharing, soft/hard task assignment, dynamic weight adjustment)