RefineDetLite: A Lightweight One-stage Object Detection Framework for CPU-only Devices (2019-11-20T00:00:00.000000Z)

TL;DR

A lightweight residual-like backbone with large receptive fields and wide dimensions for low-level features, which are crucial for detection tasks is proposed, and a light-head detection part is designed to match the backbone capability.

Abstract

Previous state-of-the-art real-time object detectors have been reported on GPUs which are extremely expensive for processing massive data and in resource-restricted scenarios. Therefore, high efficiency object detectors on CPU-only devices are urgently-needed in industry. The floating-point operations (FLOPs1) of networks are not strictly proportional to the running speed on CPU devices, which inspires the design of an exactly "fast" and "accurate" object detector. After investigating the concern gaps between classification networks and detection backbones, and following the design principles of efficient networks, we propose a lightweight residual-like backbone with large receptive fields and wide dimensions for low-level features, which are crucial for detection tasks. Correspondingly, we also design a light-head detection part to match the backbone capability. Furthermore, by analyzing the drawbacks of current one-stage detector training strategies, we also propose three orthogonal training strategies—IOU-guided loss, classes-aware weighting method and balanced multitask training approach. Without bells and whistles, our proposed RefineDetLite achieves 26.8 mAP on the MSCO-CO benchmark at a speed of 130 ms/pic on a single-thread CPU. The detection accuracy can be further increased to 29.6 mAP by integrating all the proposed training strategies, without apparent speed drop.

Authors

Mengyuan Liu

3 papers

Chen Chen

1 papers

Xiandong Meng

1 papers

TL;DR

Abstract

Authors

References53 items

Distance-IoU Loss: Faster and Better Learning for Bounding Box Regression

Searching for MobileNetV3

NAS-FPN: Learning Scalable Feature Pyramid Architecture for Object Detection

Res2Net: A New Multi-Scale Backbone Architecture

Precise Detection in Densely Packed Scenes

ThunderNet: Towards Real-Time Generic Object Detection on Mobile Devices

Generalized Intersection Over Union: A Metric and a Loss for Bounding Box Regression

Consistent Optimization for Single-Shot Object Detection

Gradient Harmonized Single-stage Detector

Bounding Box Regression With Uncertainty for Accurate Object Detection

Context Refinement for Object Detection

ShuffleNet V2: Practical Guidelines for Efficient CNN Architecture Design

Tiny-DSOD: Lightweight Object Detection for Resource-Restricted Usages

Improving Image Captioning with Conditional Generative Adversarial Nets

Pelee: A Real-Time Object Detection System on Mobile Devices

DetNet: A Backbone network for Object Detection

YOLOv3: An Incremental Improvement

SqueezeNext: Hardware-Aware Neural Network Design

Tiny SSD: A Tiny Single-Shot Detection Deep Convolutional Neural Network for Real-Time Embedded Object Detection

MobileNetV2: Inverted Residuals and Linear Bottlenecks

FSSD: Feature Fusion Single Shot Multibox Detector

Relation Networks for Object Detection

Light-Head R-CNN: In Defense of Two-Stage Object Detector

Single-Shot Refinement Neural Network for Object Detection

Focal Loss for Dense Object Detection

DSOD: Learning Deeply Supervised Object Detectors from Scratch

Bottom-Up and Top-Down Attention for Image Captioning and Visual Question Answering

ShuffleNet: An Extremely Efficient Convolutional Neural Network for Mobile Devices

Multi-task Learning Using Uncertainty to Weigh Losses for Scene Geometry and Semantics

MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications

Soft-NMS — Improving Object Detection with One Line of Code

A-Fast-RCNN: Hard Positive Generation via Adversary for Object Detection

DSSD : Deconvolutional Single Shot Detector

YOLO9000: Better, Faster, Stronger

Beyond Skip Connections: Top-Down Modulation for Object Detection

Feature Pyramid Networks for Object Detection

SqueezeDet: Unified, Small, Low Power Fully Convolutional Neural Networks for Real-Time Object Detection for Autonomous Driving

POI: Multiple Object Tracking with High Performance Detection and Appearance Feature

UnitBox: An Advanced Object Detection Network

A Unified Multi-scale Deep Convolutional Neural Network for Fast Object Detection

Training Region-Based Object Detectors with Online Hard Example Mining

HyperNet: Towards Accurate Region Proposal Generation and Joint Object Detection

SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <1MB model size

G-CNN: An Iterative Grid Based Object Detector

Inside-Outside Net: Detecting Objects in Context with Skip Pooling and Recurrent Neural Networks

Deep Residual Learning for Image Recognition

SSD: Single Shot MultiBox Detector

You Only Look Once: Unified, Real-Time Object Detection

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

Fast R-CNN

Very Deep Convolutional Networks for Large-Scale Image Recognition

Microsoft COCO: Common Objects in Context

Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation

Field of Study

Journal Information

Name

Page