Frustratingly Simple Few-Shot Object Detection (2020-03-16T00:00:00.000000Z)

TL;DR

This work finds that fine-tuning only the last layer of existing detectors on rare classes is crucial to the few-shot object detection task, and establishes a new state of the art on the revised benchmarks.

Abstract

Detecting rare objects from a few examples is an emerging problem. Prior works show meta-learning is a promising approach. But, fine-tuning techniques have drawn scant attention. We find that fine-tuning only the last layer of existing detectors on rare classes is crucial to the few-shot object detection task. Such a simple approach outperforms the meta-learning methods by roughly 2~20 points on current benchmarks and sometimes even doubles the accuracy of the prior methods. However, the high variance in the few samples often leads to the unreliability of existing benchmarks. We revise the evaluation protocols by sampling multiple groups of training examples to obtain stable comparisons and build new benchmarks based on three datasets: PASCAL VOC, COCO and LVIS. Again, our fine-tuning approach establishes a new state of the art on the revised benchmarks. The code as well as the pretrained models are available at this https URL.

Authors

Trevor Darrell

47 papers

Joseph E. Gonzalez

10 papers

F. Yu

18 papers

TL;DR

Abstract

Authors

References31 items

Meta-Learning to Detect Rare Objects

Meta R-CNN: Towards General Solver for Instance-Level Low-Shot Learning

A Baseline for Few-Shot Image Classification

LVIS: A Dataset for Large Vocabulary Instance Segmentation

TAFE-Net: Task-Aware Feature Embeddings for Low Shot Learning

A Closer Look at Few-shot Classification

Few-Shot Object Detection via Feature Reweighting

Meta-Learning with Latent Embedding Optimization

Dynamic Few-Shot Visual Learning Without Forgetting

On First-Order Meta-Learning Algorithms

Low-Shot Learning with Imprinted Weights

Prototypical Networks for Few-shot Learning

Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks

Feature Pyramid Networks for Object Detection

Matching Networks for One Shot Learning

Low-Shot Visual Recognition by Shrinking and Hallucinating Features

Deep Residual Learning for Image Recognition

Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks

Very Deep Convolutional Networks for Large-Scale Image Recognition

Microsoft COCO: Common Objects in Context

They call it like they see it: spontaneous naming and attention to shape.

Object name Learning Provides On-the-Job Training for Attention

The importance of shape in early lexical learning

2019b) shows that there are only about 2 points of difference when using a one or two-stage detector

Siamese Neural Networks for One-Shot Image Recognition

The PASCAL Visual Object Classes Challenge

The Pascal Visual Object Classes Challenge 2006 ( VOC 2006 ) Results

Frustratingly Simple Few-Shot Object Detection

We observe that using a cosine similarity based classiﬁer can achieve much higher accuracy on base classes especially in higher shots

differences are less signiﬁcant than on PASCAL VOC, similar observations can be made about accuracy on base classes and novel classes

We plot the model performance measured by AP, AP50 and AP75 of up to 40 random groups of training shots across all three splits in Figure 7

Field of Study

Journal Information

Name

Volume

Venue Information

Name

Type

URL

Alternate Names