Dynamic Clone Transformer for Efficient Convolutional Neural Netwoks
This work addresses model deployment challenges on resource-constrained platforms, presenting an incremental improvement in network design.
The paper tackles the performance-efficiency trade-off in convolutional neural networks (ConvNets) for vision tasks by introducing a dynamic clone transformer (DCT) module, which achieves efficient channel expansion with minimal computational cost, though no specific numerical results are provided.
Convolutional networks (ConvNets) have shown impressive capability to solve various vision tasks. Nevertheless, the trade-off between performance and efficiency is still a challenge for a feasible model deployment on resource-constrained platforms. In this paper, we introduce a novel concept termed multi-path fully connected pattern (MPFC) to rethink the interdependencies of topology pattern, accuracy and efficiency for ConvNets. Inspired by MPFC, we further propose a dual-branch module named dynamic clone transformer (DCT) where one branch generates multiple replicas from inputs and another branch reforms those clones through a series of difference vectors conditional on inputs itself to produce more variants. This operation allows the self-expansion of channel-wise information in a data-driven way with little computational cost while providing sufficient learning capacity, which is a potential unit to replace computationally expensive pointwise convolution as an expansion layer in the bottleneck structure.