With the rapid development of artificial intelligence, unmanned systems such as autonomous driving and embodied intelligence are constantly being promoted and applied in real-life scenarios.
As the core means of obtaining information, visual perception plays a very important role in these intelligent systems.
However, how to achieve efficient, accurate and robust visual perception in a dynamic, diverse and unpredictable environment is still a difficult problem to be solved.
(Photo source: Tsinghua University) In open world scenarios, intelligent systems not only have to process large amounts of data, but also deal with various extreme events, such as sudden dangers, violent light changes at tunnel entrances, and strong driving scenarios at night.
Flash interference, etc.
Traditional visual perception chips are limited by “power consumption walls” and “bandwidth walls”, and often face distortion, failure or high latency problems when processing these scenarios, which seriously affects the stability and security of the system.
Return to the first electric network home page>,.
