Path Aggregation Network for Instance Segmentation

By Shu Li et al

Published on Sept. 18, 2018

Read the original document by opening this link in a new tab.

Table of Contents

1. Introduction
2. Related Work
3. Framework
3.1. Bottom-up Path Augmentation
3.2. Adaptive Feature Pooling
3.3. Fully-connected Fusion
4. Experimental Results
5. Conclusion

Summary

The document presents the Path Aggregation Network (PANet) for boosting information flow in instance segmentation. PANet enhances the feature hierarchy with accurate localization signals, introduces adaptive feature pooling, and a complementary branch for improved mask prediction. The paper discusses the importance of information propagation in neural networks and details the design principles of PANet. Experimental results show PANet's state-of-the-art performance on various datasets and tasks. The approach is shared by object detection and instance segmentation, leading to enhanced performance in both areas.