Abstract:Learning pyramidal feature representations is crucial for recognizing object instances at different scales. Feature Pyramid Network (FPN) is the classic architecture to build a feature pyramid with high-level semantics throughout. However, intrinsic defects in feature extraction and fusion inhibit FPN from further aggregating more discriminative features. In this work, we propose Attention Aggregation based Feature Pyramid Network (A 2 -FPN), to improve multi-scale feature learning through attention-guided feature aggregation. In feature extraction, it extracts discriminative features by collecting-distributing multi-level global context features, and mitigates the semantic information loss due to drastically reduced channels. In feature fusion, it aggregates complementary information from adjacent features to generate location-wise reassembly kernels for content-aware sampling, and employs channel-wise reweighting to enhance the semantic consistency before element-wise addition. A 2 -FPN shows consistent gains on different instance segmentation frameworks. By replacing FPN with A 2 -FPN in Mask R-CNN, our model boosts the performance by 2.1% and 1.6% mask AP when using ResNet-50 and ResNet-101 as backbone, respectively. Moreover, A 2 -FPN achieves an improvement of 2.0% and 1.4% mask AP when integrated into the strong baselines such as Cascade Mask R-CNN and Hybrid Task Cascade.

FPANet: Feature-enhanced Position Attention Network for Semantic Segmentation

SAFPN: a Full Semantic Feature Pyramid Network for Object Detection

Deep Dual-Stream Network with Scale Context Selection Attention Module for Semantic Segmentation

FPANet: Feature Pyramid Aggregation Network for Real-Time Semantic Segmentation

SemSegFPA: Semantic Segmentation Feature Pyramid Network with Attention Module

PPNet : Pooling Position Attention Network for Semantic Segmentation

Using Features Specifically: an Efficient Network for Scene Segmentation Based on Dedicated Attention Mechanisms

FE-Net: Feature Enhancement Segmentation Network

Fully Attentional Network for Semantic Segmentation

Cross Attention Network For Semantic Segmentation

Position Attention Optimized Deep Semantic Segmentation

Fully Convolutional Neural Networks With Full-Scale-Features For Semantic Segmentation

FPAN: Fine-grained and Progressive Attention Localization Network for Data Retrieval.

DPANET:Dual Pooling Attention Network for Semantic Segmentation

Dual Path Attention Net For Remote Sensing Semantic Image Segmentation

Fully convolutional attention network for biomedical image segmentation

Learning a Discriminative Feature Network for Semantic Segmentation

Enhanced Feature Pyramid Network for Semantic Segmentation.

Point Attention Network for Point Cloud Semantic Segmentation.

A<sup>2</sup>-FPN: Attention Aggregation based Feature Pyramid Network for Instance Segmentation

Attention Guided Global Enhancement and Local Refinement Network for Semantic Segmentation