Tied Block Convolution: Leaner and Better CNNs with Shared Thinner Filters

TL;DR

Tied Block Convolution (TBC) introduces shared thinner filters across channel blocks in CNNs, reducing redundancy, improving efficiency, and enhancing performance in classification, detection, and segmentation tasks, especially under occlusion.

Contribution

The paper proposes Tied Block Convolution, a novel filter sharing method that improves CNN efficiency and accuracy across multiple tasks and can be extended to various network components.

Findings

TBC achieves significant accuracy gains over standard convolution.

TBC reduces parameter count, exemplified by TiedSE using 64x fewer parameters.

TBC improves object detection AP by 6% under high occlusion conditions.

Abstract

Convolution is the main building block of convolutional neural networks (CNN). We observe that an optimized CNN often has highly correlated filters as the number of channels increases with depth, reducing the expressive power of feature representations. We propose Tied Block Convolution (TBC) that shares the same thinner filters over equal blocks of channels and produces multiple responses with a single filter. The concept of TBC can also be extended to group convolution and fully connected layers, and can be applied to various backbone networks and attention modules. Our extensive experimentation on classification, detection, instance segmentation, and attention demonstrates TBC's significant across-the-board gain over standard convolution and group convolution. The proposed TiedSE attention module can even use 64 times fewer parameters than the SE module to achieve comparable performance. In particular, standard CNNs often fail to accurately aggregate information in the presence of occlusion and result in multiple redundant partial object proposals. By sharing filters across channels, TBC reduces correlation and can effectively handle highly overlapping instances. TBC increases the average precision for object detection on MS-COCO by 6% when the occlusion ratio is 80%. Our code will be released.