Recalibrating Fully Convolutional Networks With Spatial and Channel “Squeeze and Excitation” Blocks

By Abhijit Guha Roy, Nassir Navab, Christian Wachinger. 2018

467 citations recorded by the source corpus.

In a wide range of semantic segmentation tasks, fully convolutional neural networks (F-CNNs) have been successfully leveraged to achieve the state-of-the-art performance. Architectural innovations of F-CNNs have mainly been on improving spatial encoding or network connectivity to aid gradient flow. In this paper, we aim toward an alternate direction of recalibrating the learned feature maps adaptively, boosting meaningful features while suppressing weak ones. The recalibration is achieved by simple computational blocks that can be easily integrated in F-CNNs architectures. We draw our inspiration from the recently proposed “squeeze and excitation” (SE) modules for channel recalibration for image classification. Toward this end, we introduce three variants of SE modules for segmentation: 1) squeezing spatially and exciting channel wise; 2) squeezing channel wise and exciting spatially; and 3) joint spatial and channel SE. We effectively incorporate the proposed SE blocks in three state-of-the-art F-CNNs and demonstrate a consistent improvement of segmentation accuracy on three challenging benchmark datasets. Importantly, SE blocks only lead to a minimal increase in model complexity of about 1.5%, while the Dice score increases by 4%–9% in the case of U-Net. Hence, we believe that SE blocks can be an integral part of future F-CNN architectures.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

Citations in the corpus

Continual Barlow Twins: Continual Self-Supervised Learning for Remote Sensing Semantic Segmentation (2022, methodology): “use UNet++ [9], with the Squeeze and Excitation strategy [27]”
FiBiNET: combining feature importance and bilinear feature interaction for click-through rate prediction (2019, methodology): “[21] introduces three variants of the SE modules for semantic segmentation task.”
Attention mechanisms in computer vision: A survey (2021, methodology): “[108] proposed spatial and channel SE blocks (scSE).”
ECA-Net: Efficient Channel Attention for Deep Convolutional Neural Networks (2019, methodology): “GE (Hu et al. 2018) explores spatial extension using a depth-wise convolution (Chollet 2017) to aggregate features. scSE (Roy, Navab, and Wachinger 2019) and CBAM (Woo et al. 2018) Obviously, all of the above methods focus on developing sophisticated attention modules for better performance.”
FcaNet: Frequency Channel Attention Networks (2020, methodology): “Inspired by these works, a series of works like BAM [28], DAN [9], CBAM [39], scSE [32], and CoordAtten-tion [19] are proposed to fuse spatial attention [45] and channel attention.”
A Noise-Robust Framework for Automatic Segmentation of COVID-19 Pneumonia Lesions From CT Images (2020, background): “…of nnU-Net [41]; 2) Attention U-Net [36] that uses a spatial attention gate to enable the network to focus more on the target region; 3) ScSE U-Net [46] that combines concurrent Spatial and Channel SE (ScSE) blocks with U-Net [16]; 4) ESPNetv2 [47] that is a light-weight, power efﬁcient and…”