Multi-Scale Weight Sharing Network for Image Recognition
share
In this paper, we explore the idea of weight sharing over multiple scales in convolutional networks. Inspired by traditional computer vision approaches, we share the weights of convolution kernels over different scales in the same layers of the network. Although multi-scale feature aggregation and sharing inside convolutional networks are common in practice, none of the previous works address the issue of convolutional weight sharing. We evaluate our weight sharing scheme on two heterogeneous image recognition datasets - ImageNet (object recognition) and Places365-Standard (scene classification). With approximately 25 similar performance compared to baseline ResNets. Shared-weight models are further validated via transfer learning experiments on four additional image recognition datasets - Caltech256 and Stanford 40 Actions (object-centric) and SUN397 and MIT Inddor67 (scene-centric). Experimental results demonstrate significant redundancy in the vanilla implementations of the deeper networks, and also indicate that a shift towards increasing the receptive field per parameter may improve future convolutional network architectures.
READ FULL TEXT