Dynamically Reconfigurable Perception using Dataflow Parameterization of Channel Attention

Despite the increasing performance of deep learning models for intelligent perception tasks, the size and complex-ity of state-of-the-art models have become increasingly large, which makes them difficult to deploy in resource-constrained edge computing environments. Moreover, the conditions and constraints under which these models operate may be im-possible to fully anticipate at design time, and may undergo significant changes during inference. In this paper, we address these challenges by developing a new parameterized design approach for image-based perception that enables efficient and dynamic reconfiguration of convolutions using channel attention. Compared to switching among sets of multiple complete neural network models, the proposed reconfiguration approach is much more streamlined in terms of resource requirements, while providing a high level of adaptability to handle unpredictable and dynamically-varying operational scenarios. The efficiency and adaptability of the proposed approach are demonstrated on ResNetSO integrated with the Convolutional Block Attention Module (CBAM) concept.