Fuse Local and Global Semantics in Representation Learning

TL;DR

This paper introduces FLAGS, a method that combines local and global semantic features to create richer image representations, improving transferability and performance in downstream tasks like detection and segmentation.

Contribution

FLAGS is a novel approach that effectively fuses local and global semantics in image representations, enhancing transferability and downstream task performance.

Findings

Promising results under linear evaluation protocol.

Effective transferability demonstrated on PASCAL VOC and COCO.

Improved performance in detection and segmentation tasks.

Abstract

We propose Fuse Local and Global Semantics in Representation Learning (FLAGS) to generate richer representations. FLAGS aims at extract both global and local semantics from images to benefit various downstream tasks. It shows promising results under common linear evaluation protocol. We also conduct detection and segmentation on PASCAL VOC and COCO to show the representations extracted by FLAGS are transferable.