计算机视觉工程师
Computer Vision Engineer
4Data Limited
岗位信息
以下为根据雇主原文整理的中文岗位要点，方便快速了解职责与要求。完整条件请展开下方招聘原文核对。
岗位概况
利用平台合成数据提升智慧城市场景的真实计算机视觉效果，设计并运行受控实验，搭建可复现的训练与评测流程，建立仅真实、仅合成与混合数据基线，刻画并缩小合成到真实的差距，评估目标检测与立体视觉等模型。
岗位要点
- 搭建可复现的训练与评测流程，建立仅真实、仅合成与混合数据基线
- 通过改变环境条件与稀有事件频率，刻画合成到真实的差距
- 对比数据比例、真实标注预算与针对性极端场景，研究合成数据适配与预训练方法
- 评估计算机视觉模型并分析误差，记录实验并撰写技术报告
任职要求
- 具备计算、数据分析、计算机视觉或机器学习背景，并有训练模型的实操经验
- 能实现并对比目标检测、立体视觉或相近感知任务的训练与评测流程
- 注重实验控制，包括数据集划分、匹配的模型容量与算力预算
申请截止：2027-01-14（以雇主原公告为准）
查看完整招聘原文
Computer Vision Engineer
We are looking for a Computer Vision Engineer to use synthetic data from our platform to improve real-world computer vision for smart-city applications. You will design and run controlled experiments on when synthetic data helps, and how training methods and model components can narrow the gap between rendered scenes and real sensors.
This is a hands-on research role for someone who enjoys experimental machine learning, careful baselines, and turning model failures into specific requirements for how synthetic scenes should be generated.
What You’ll Work On
- Build reproducible training and evaluation pipelines, and establish real-only, synthetic-only, and mixed-data baselines.
- Characterise the synthetic-to-real gap by varying environment conditions, and rare-event frequency.
- Compare data ratios, real-label budgets, and targeted corner cases against real-only training and naïve pooling.
- Investigate synthetic data adaptation methods to improve models’ performance.
- Study pretraining approach that leverages synthetic data.
- Evaluate computer vision models, including repeated runs on ordinary and challenging conditions.
- Analyze errors and document the work in experiment records and technical reports.
What We’re Looking For
- A background in computing, data analytics, computer vision, or machine learning, with hands-on experience training models.
- Ability to implement and compare training and evaluation pipelines for object detection, stereo vision, or a closely related perception task.
- Care with experimental controls, including dataset splits, matched model capacity, compute budgets, and repeated training runs.
- Comfort reading current research, selecting baselines, and interpreting negative transfer rather than only pursuing a higher score.
- Familiarity with quantitative evaluation across ordinary conditions and difficult cases such as weather, occlusion, scale, and rare events.
- Interest in synthetic data, domain adaptation, and how rendered 3D scenes are used to train real-world AI.
- Clear writing for experiment records, ablation reports, and recommendations that a data-generation team can act on.
- Comfort working in an early-stage R&D environment where the data recipe and model design change as evidence accumulates.
Nice to Have
- Project, internship, or work experience in computer vision, deep learning, or applied AI.
- Familiarity with domain randomisation, domain adaptation, representation learning, or training with limited labelled data.
- Experience with depth, disparity, stereo correspondence, or multi-view geometry.
- Experience training on both synthetic and real images and diagnosing why a model fails to transfer.
申请截止：2027-01-14（以雇主原公告为准）