Gated Channel Transformation for Visual Recognition
This work addresses the need for efficient and interpretable channel transformation in visual recognition tasks, offering a domain-specific improvement over existing methods like Squeeze-and-Excitation Networks.
The paper tackles the problem of modeling channel relationships in deep convolutional neural networks for visual recognition by proposing a lightweight transformation unit that uses explainable control variables to determine neuron competition or cooperation, achieving clear performance margins on ImageNet, COCO, and Kinetics datasets.
In this work, we propose a generally applicable transformation unit for visual recognition with deep convolutional neural networks. This transformation explicitly models channel relationships with explainable control variables. These variables determine the neuron behaviors of competition or cooperation, and they are jointly optimized with the convolutional weight towards more accurate recognition. In Squeeze-and-Excitation (SE) Networks, the channel relationships are implicitly learned by fully connected layers, and the SE block is integrated at the block-level. We instead introduce a channel normalization layer to reduce the number of parameters and computational complexity. This lightweight layer incorporates a simple l2 normalization, enabling our transformation unit applicable to operator-level without much increase of additional parameters. Extensive experiments demonstrate the effectiveness of our unit with clear margins on many vision tasks, i.e., image classification on ImageNet, object detection and instance segmentation on COCO, video classification on Kinetics.