Article source: zhuanlan.zhihu.com/p/115035951

Author: Red Stone \

As an important branch of computer vision, target detection has become a hot spot of global artificial intelligence research in recent years with the deepening of neural network theory research and the significant improvement of hardware GPU computing power.

From 2013 to 2020, from r-CNN and OverFeat, to SSD, YOLO V3, and last year’s M2Det, new models have emerged one after another, and their performance is getting better and better! In this paper, 52 target detection models and their performance comparison will be completely summarized, including a complete list of literature papers.

Let’s get straight to the point and list these 52 target detection models:

This watch is too comfortable! This technical route summary of target detection is from a well-known project on GitHub by Lee Hoseong, who graduated from Seoul National University in Electrical and computer Engineering and has received 7.6 K Star.

The project address is:

Github.com/hoya012/dee…

The technology route spans a time line from 2013 to early 2020, and the figure above summarizes all representative models for target detection during this period. The parts marked in red are relatively important models that need to be mastered.

Update log

It is worth mentioning that Red Stone recommended this project as early as the beginning of last year, and the author has been updating. As of February 2020, the author’s main updates are as follows:

  • 2019.2: Update 3 papers
  • 2019.3: Update charts and code links
  • 2019.4: Update ICLR 2019 and CVPR 2019 papers
  • 2019.5: Update the CVPR 2019 paper
  • 2019.6: Update CVPR 2019 paper and data set paper
  • 2019.7: Update BMVC 2019 papers and some ICCV 2019 papers
  • 2019.9: Update NeurIPS 2019 paper and ICCV 2019 paper
  • November 2019: Update some AAAI 2020 papers and other papers
  • 2020.1: Update THE ICLR 2020 paper and other papers

Details below!

Model performance comparison table

FPS comparisons are often inaccurate due to hardware differences (e.g., CPU, GPU, RAM, etc.). A more appropriate comparison is to measure the performance of all models in the same hardware configuration. The performance comparison results of all the above models are as follows:

From the table above, we can clearly see the performance of different models on VOC07, VOC12 and COCO data sets. At the same time, the sources of model papers are listed. \

Here are some models highlighted in red for a brief introduction.

Model paper

In 2014,

R-CNN

Rich feature hierarchies for accurate object detection and semantic segmentation | [CVPR’ 14]

Paper:

Arxiv.org/pdf/1311.25…

Official code Caffe:

Github.com/rbgirshick/…

OverFeat

OverFeat: Integrated Recognition, Localization and Detection using Convolutional Networks | [ICLR’ 14]

Paper:

Arxiv.org/pdf/1312.62…

Official code Torch:

Github.com/sermanet/Ov…

In 2015,

Fast R-CNN

Fast R-CNN | [ICCV’ 15]

Paper:

Arxiv.org/pdf/1504.08…

Official code caffe:

Github.com/rbgirshick/…

Faster R-CNN

Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks | [NIPS’ 15]

Paper:

Cca shut. Nips. Cc/paper / 5638 -…

Official code caffe:

Github.com/rbgirshick/…

Unofficial code tensorflow:

Github.com/endernewton…

Unofficial code PyTorch:

Github.com/jwyang/fast…

In 2016,

OHEM

Training Region-based Object Detectors with Online Hard Example Mining | [CVPR’ 16]

Paper:

Arxiv.org/pdf/1604.03…

Official code caffe:

Github.com/abhi2610/oh…

YOLO v1

You Only Look Once: Unified, Real-Time Object Detection | [CVPR’ 16]

Paper:

Arxiv.org/pdf/1506.02…

Official code C:

Pjreddie.com/darknet/yol…

SSD

SSD: Single Shot MultiBox Detector | [ECCV’ 16]

Paper:

Arxiv.org/pdf/1512.02…

Official code caffe:

Github.com/weiliu89/ca…

Unofficial code tensorflow:

Github.com/balancap/SS…

Unofficial code PyTorch:

Github.com/amdegroot/s…

R-FCN

R-FCN: Object Detection via Region-based Fully Convolutional Networks | [NIPS’ 16]

Paper:

Arxiv.org/pdf/1605.06…

Official code caffe:

Github.com/daijifeng00…

Unofficial code caffe:

Github.com/YuwenXiong/…

In 2017,

YOLO v2

YOLO9000: Better, Faster, Stronger | [CVPR’ 17]

Paper:

Arxiv.org/pdf/1612.08…

Official code C:

Pjreddie.com/darknet/yol…

Unofficial code caffe:

Github.com/quhezheng/c…

Unofficial code tensorflow:

Github.com/nilboy/tens…

Unofficial code tensorflow:

Github.com/sualab/obje…

Unofficial code PyTorch:

Github.com/longcw/yolo…

FPN

Feature Pyramid Networks for Object Detection | [CVPR’ 17]

Paper:

Openaccess.thecvf.com/content_cvp…

Unofficial code caffe:

github.com/unsky/FPN

RetinaNet

Focal Loss for Dense Object Detection | [ICCV’ 17]

Paper:

Arxiv.org/pdf/1708.02…

Official code keras:

Github.com/fizyr/keras…

Unofficial code PyTorch:

Github.com/kuangliu/py…

Unofficial code mxnet:

Github.com/unsky/Retin…

Unofficial code tensorflow:

Github.com/tensorflow/…

Mask R-CNN

Mask R-CNN | [ICCV’ 17]

Paper:

Openaccess.thecvf.com/content_ICC…

Caffe2:

Github.com/facebookres…

Unofficial code tensorflow:

Github.com/matterport/…

Unofficial code tensorflow:

Github.com/CharlesShan…

Unofficial code PyTorch:

Github.com/multimodall…

In 2018,

YOLO v3

YOLOv3: An Incremental Improvement | [arXiv’ 18]

Paper:

Pjreddie.com/media/files…

Official code C:

Pjreddie.com/darknet/yol…

Unofficial code PyTorch:

Github.com/ayooshkathu…

Unofficial code PyTorch:

Github.com/eriklindern…

Unofficial code Keras:

Github.com/qqwweee/ker…

Unofficial code tensorflow:

Github.com/mystic123/t…

RefineDet

Single-Shot Refinement Neural Network for Object Detection | [CVPR’ 18]

Paper:

Openaccess.thecvf.com/content_cvp…

Official code caffe:

Github.com/sfzhang15/R…

Unofficial code Chainer:

Github.com/fukatani/Re…

Unofficial code PyTorch:

Github.com/lzx1413/Pyt…

In 2019,

M2Det

M2Det: A Single-Shot Object Detector based on Multi-Level Feature Pyramid Network | [AAAI’ 19]

Paper:

Arxiv.org/pdf/1811.04…

Official code PyTorch:

Github.com/qijiezhao/M…

In 2020,

Spiking-YOLO

Spiking-YOLO: Spiking Neural Network for Real-time Object Detection | [AAAI’ 20]

Paper:

Arxiv.org/pdf/1903.06…

Data set papers

At the same time, the author also lists the public data sets commonly used by the above models: VOC, ILSVRC, COCO, as shown in the following table:

The data set used for target detection is as follows:

The above is a summary of 52 deep learning detection models. This project can be said to be a good summary of the target detection model in recent years, including papers and source code. Hope to help you!

Machine Learning Online Manual Deep Learning online Manual AI Basics download (PDF updated to25Set) site QQ group1003271085To join the wechat group, please reply to "add group" to get a discount station knowledge planet coupon, please reply to "knowledge planet" like the article, click on itCopy the code