One-Shot Object Detection with Co-Attention and Co-Excitation

TL;DR

This paper introduces a novel co-attention and co-excitation framework for one-shot object detection, enabling the detection of unseen classes by leveraging non-local operations, adaptive feature emphasis, and a ranking loss.

Contribution

It proposes a new framework combining co-attention, co-excitation, and a ranking loss for effective one-shot object detection of unseen classes.

Findings

Achieves strong baseline performance on VOC and MS-COCO datasets.

Effectively detects objects from both seen and unseen classes.

Introduces a novel combination of non-local operations and adaptive feature emphasis.

Abstract

This paper aims to tackle the challenging problem of one-shot object detection. Given a query image patch whose class label is not included in the training data, the goal of the task is to detect all instances of the same class in a target image. To this end, we develop a novel {\em co-attention and co-excitation} (CoAE) framework that makes contributions in three key technical aspects. First, we propose to use the non-local operation to explore the co-attention embodied in each query-target pair and yield region proposals accounting for the one-shot situation. Second, we formulate a squeeze-and-co-excitation scheme that can adaptively emphasize correlated feature channels to help uncover relevant proposals and eventually the target objects. Third, we design a margin-based ranking loss for implicitly learning a metric to predict the similarity of a region proposal to the underlying query, no matter its class label is seen or unseen in training. The resulting model is therefore a two-stage detector that yields a strong baseline on both VOC and MS-COCO under one-shot setting of detecting objects from both seen and never-seen classes. Codes are available at https://github.com/timy90022/One-Shot-Object-Detection.