A large number of high-level algorithms have been incubated for perception tasks such as unknown objects in real environments, which are mainly divided into online processing and offline processing:
Unknown task identification during online runtime, identifying new categories or new domains to trigger retraining or sending text messages, etc. Unknown task processing during online runtime, identifying new tasks and combining multiple edge and cloud models to provide real-time backup, rather than just sending text messages and waiting for offline training.
Offline training for unknown task processing, using GAN to generate kuwait mobile phone number list data for unknown tasks, while using self-supervision methods to reduce annotation costs.
In 2023, we will also optimize the unknown task processing of online runtime and offline training in combination with large models. The relevant data sets and baseline algorithm processes have been open sourced in KubeEdge Ianvs.
Figure 13 High-level open source algorithm for robot intelligent navigation
Cloud-native edge intelligence is crossing the boundaries of a new era. Change is already here, and we hope to work together with colleagues from industry, academia, and research to move towards the future. If you want to learn more about cloud-native edge computing, the KubeEdge open source community and the monograph "Edge Computing Theory and System Practice: Implementation Based on CNCF KubeEdge" may provide you with more information.