Inception Convolution with Efficient Dilation Search
This work offers an incremental improvement in computer vision tasks for practitioners by enhancing the performance of existing models through a novel convolution variant.
This paper proposes inception convolution, a new type of dilated convolution with independent dilation patterns across axes, channels, and layers. Coupled with an efficient dilation optimization (EDO) search algorithm, it significantly improves the AP of Faster R-CNN on MS COCO from 36.4% to 39.2% by replacing standard convolutions in ResNet-50.
As a variant of standard convolution, a dilated convolution can control effective receptive fields and handle large scale variance of objects without introducing additional computational costs. To fully explore the potential of dilated convolution, we proposed a new type of dilated convolution (referred to as inception convolution), where the convolution operations have independent dilation patterns among different axes, channels and layers. To develop a practical method for learning complex inception convolution based on the data, a simple but effective search algorithm, referred to as efficient dilation optimization (EDO), is developed. Based on statistical optimization, the EDO method operates in a low-cost manner and is extremely fast when it is applied on large scale datasets. Empirical results validate that our method achieves consistent performance gains for image recognition, object detection, instance segmentation, human detection, and human pose estimation. For instance, by simply replacing the 3x3 standard convolution in the ResNet-50 backbone with inception convolution, we significantly improve the AP of Faster R-CNN from 36.4% to 39.2% on MS COCO.