数据依然是王道：使用已有成熟模型加工、生成数据是一条低成本高收益的路径。N3D-VLM巧妙地利用现有2D数据配合 Metric3D 进行“升维”构建3D训练集，这种思想在其他领域或许同样适用。

但如果结合小鹏自动驾驶负责人刘先明的访谈内容来看，这并非一次简单的技术摇摆，而更像是在当前算力、数据与工程约束下，对数据扩展效率（data scaling efficiency，数据规模扩展效率）和系统瓶颈的重新权衡。

Milestone Systems, a provider of data-driven video technology, has released an advanced vision language model (VLM) ...

Vision language models trained on traffic data help cities and transport networks move from reactive video monitoring to ...

Milestone announced the traffic-focused VLM, powered by NVIDIA Cosmos Reason, supports automated video summarization in ...

There are different types of AI models available in the market for users to choose from, and it will largely depend on the type of service they need from the machine learning technology, and Google ...

As I highlighted in my last article, two decades after the DARPA Grand Challenge, the autonomous vehicle (AV) industry is still waiting for breakthroughs—particularly in addressing the “long tail ...

Linker Vision's collaboration with NVIDIA was a key highlight of the showcase. The Observ platform's video analytic system integrates with the NVIDIA AI Blueprint for video search and summarization, ...

MIT researchers discovered that vision-language models often fail to understand negation, ignoring words like “not” or “without.” This flaw can flip diagnoses or decisions, with models sometimes ...

Just when you thought the pace of change of AI models couldn’t get any faster, it accelerates yet again. In the popular news media, the introduction of DeepSeek in January 2025 created a moment that ...