Dual Complementary Dynamic Convolution for Image Recognition
This work addresses limitations in convolutional neural networks for computer vision tasks, offering an incremental improvement over existing dynamic convolutions.
The paper tackles the sample and content agnostic problems in vanilla convolution by proposing a dual complementary dynamic convolution (DCDC) operator that models scene features as a combination of local spatial-adaptive and global shift-invariant parts, resulting in DCDC-ResNets significantly outperforming vanilla ResNets and most state-of-the-art dynamic convolutional networks on image classification and downstream tasks with lower FLOPs and parameters.
As a powerful engine, vanilla convolution has promoted huge breakthroughs in various computer tasks. However, it often suffers from sample and content agnostic problems, which limits the representation capacities of the convolutional neural networks (CNNs). In this paper, we for the first time model the scene features as a combination of the local spatial-adaptive parts owned by the individual and the global shift-invariant parts shared to all individuals, and then propose a novel two-branch dual complementary dynamic convolution (DCDC) operator to flexibly deal with these two types of features. The DCDC operator overcomes the limitations of vanilla convolution and most existing dynamic convolutions who capture only spatial-adaptive features, and thus markedly boosts the representation capacities of CNNs. Experiments show that the DCDC operator based ResNets (DCDC-ResNets) significantly outperform vanilla ResNets and most state-of-the-art dynamic convolutional networks on image classification, as well as downstream tasks including object detection, instance and panoptic segmentation tasks, while with lower FLOPs and parameters.