Efficient Transmission Latency Constraint in Deep Reinforcement Learning for Zero-Shot Object Recognition

The article begins by introducing the challenges of mobile AR object detection, where the sheer volume of collected data leads to impractical communication overhead. To address this issue, the authors propose incorporating a SKB into the system, which enables the optimization of multi-level feature transmission. The SKB is constructed by annotating semantic vectors for each class of objects and their attributes, providing a comprehensive understanding of the relationships between them.
The proposed method involves dividing the input data into multiple levels based on their importance and transmitting only the necessary features at each level. This approach reduces the communication overhead while maintaining accurate object detection. The authors constrain the transmission latency to ensure timely processing and optimize the SKB size to balance accuracy and efficiency.
Through simulations, the authors evaluate the proposed method and demonstrate its effectiveness in minimizing semantic loss while maintaining accurate object detection. They also compare their approach with traditional methods and show improved performance.
In summary, the article proposes a novel approach to mobile AR object detection that leverages a SKB to optimize multi-level feature transmission. By constraining transmission latency and optimizing SKB size, the proposed method minimizes semantic loss while maintaining accurate object detection. The approach is evaluated through simulations and demonstrates improved performance compared to traditional methods.