Modern Convolutional
Object Detectors
Kwanghee Choi
Sogang Univ.
Contents
- Prior Knowledge
- Object Detection
- R-CNN
- History
- Meta-architectures
- Faster R-CNN
- R-FCN
- YOLO
- Object Detection Benchmarks
- Accuracy
- Time
- Memory
Prior Knowledge: Neural Network
Ref. Fei-Fei Li & Andrej Karpathy & Justin Johnson, CS231n: Convolutional Neural Networks for Visual Recognition, Stanford Univ.
Prior Knowledge: Convolution
Ref. Aaditya Prakash, One by One Convolution: counter-intuitively useful, iamaaditya.github.io/2016/03/one-by-one-convolution
Prior Knowledge: Convolutional Neural Network
Ref. Fei-Fei Li & Andrej Karpathy & Justin Johnson, CS231n: Convolutional Neural Networks for Visual Recognition, Stanford Univ.
Prior Knowledge: Classification vs. Regression
Ref. Cyrille Rossant, IPython Interactive Computing and Visualization Cookbook, Packt Publishing
Prior Knowledge: Computer Vision Tasks
Ref. Fei-Fei Li & Andrej Karpathy & Justin Johnson, CS231n: Convolutional Neural Networks for Visual Recognition, Stanford Univ.
Object Detection: Demo
Ref. Joseph Redmon, YOLO v2, https://www.youtube.com/watch?v=VOC3huqHrss
Object Detection: History
Ref. Sam Albanie, R-FCN: Region-based Fully Convolutional Networks, VGG Reading Group
R-CNN: Pipeline
Ref. Ross Girshick et al., Rich feature hierarchies for accurate object detection and semantic segmentation, arXiv:1311.2524v5 [cs.CV] 22 Oct 2014
R-CNN: Selective Search
Ref. J. R. R. Uijlings et al., Selective Search for Object Recognition, IJCV 2013
Object Detection: History
Ref. Sam Albanie, R-FCN: Region-based Fully Convolutional Networks, VGG Reading Group
Object Detection: Meta-architectures
Ref. Jonathan Huang et al., Speed/accuracy trade-offs for modern convolutional object detectors, arXiv:1611.10012v3 [cs.CV] 25 Apr 2017
Faster R-CNN: Architecture
Ref. Shaoqing Ren et al., Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks, arXiv:1506.01497v3 [cs.CV] 6 Jan 2016
Faster R-CNN: Region Proposal Network
Ref. Shaoqing Ren et al., Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks, arXiv:1506.01497v3 [cs.CV] 6 Jan 2016
Object Detection: Meta-architectures
Ref. Jonathan Huang et al., Speed/accuracy trade-offs for modern convolutional object detectors, arXiv:1611.10012v3 [cs.CV] 25 Apr 2017
R-FCN: Architecture
Ref. Jifeng Dai et al., R-FCN: Object Detection via Region-based Fully Convolutional Network, arXiv:1605.06409v2 [cs.CV] 21 Jun 2016
R-FCN: Position-Sensitive Score Maps
Ref. Jifeng Dai et al., R-FCN: Object Detection via Region-based Fully Convolutional Network, arXiv:1605.06409v2 [cs.CV] 21 Jun 2016
Object Detection: Meta-architectures
Ref. Jonathan Huang et al., Speed/accuracy trade-offs for modern convolutional object detectors, arXiv:1611.10012v3 [cs.CV] 25 Apr 2017
YOLO: Architecture
Ref. Joseph Redmond et al., You Only Look Once: Unified, Real-Time Object Detection, arXiv:1506.02640v5 [cs.CV] 9 May 2016
YOLO: Regression Model
Ref. Joseph Redmond et al., You Only Look Once: Unified, Real-Time Object Detection, arXiv:1506.02640v5 [cs.CV] 9 May 2016
Object Detection: Meta-architectures
Ref. Jonathan Huang et al., Speed/accuracy trade-offs for modern convolutional object detectors, arXiv:1611.10012v3 [cs.CV] 25 Apr 2017
Object Detection Benchmarks: Accuracy
Ref. Jonathan Huang et al., Speed/accuracy trade-offs for modern convolutional object detectors, arXiv:1611.10012v3 [cs.CV] 25 Apr 2017
Object Detection Benchmarks: GPU Time
Ref. Jonathan Huang et al., Speed/accuracy trade-offs for modern convolutional object detectors, arXiv:1611.10012v3 [cs.CV] 25 Apr 2017
Object Detection Benchmarks: Memory Usage
Ref. Jonathan Huang et al., Speed/accuracy trade-offs for modern convolutional object detectors, arXiv:1611.10012v3 [cs.CV] 25 Apr 2017
Object Detection Benchmarks: Accuracy vs. Time
Ref. Jonathan Huang et al., Speed/accuracy trade-offs for modern convolutional object detectors, arXiv:1611.10012v3 [cs.CV] 25 Apr 2017
Welcome to the world where computer understands images.

Modern convolutional object detectors

  • 1.
  • 2.
    Contents - Prior Knowledge -Object Detection - R-CNN - History - Meta-architectures - Faster R-CNN - R-FCN - YOLO - Object Detection Benchmarks - Accuracy - Time - Memory
  • 3.
    Prior Knowledge: NeuralNetwork Ref. Fei-Fei Li & Andrej Karpathy & Justin Johnson, CS231n: Convolutional Neural Networks for Visual Recognition, Stanford Univ.
  • 4.
    Prior Knowledge: Convolution Ref.Aaditya Prakash, One by One Convolution: counter-intuitively useful, iamaaditya.github.io/2016/03/one-by-one-convolution
  • 5.
    Prior Knowledge: ConvolutionalNeural Network Ref. Fei-Fei Li & Andrej Karpathy & Justin Johnson, CS231n: Convolutional Neural Networks for Visual Recognition, Stanford Univ.
  • 6.
    Prior Knowledge: Classificationvs. Regression Ref. Cyrille Rossant, IPython Interactive Computing and Visualization Cookbook, Packt Publishing
  • 7.
    Prior Knowledge: ComputerVision Tasks Ref. Fei-Fei Li & Andrej Karpathy & Justin Johnson, CS231n: Convolutional Neural Networks for Visual Recognition, Stanford Univ.
  • 8.
    Object Detection: Demo Ref.Joseph Redmon, YOLO v2, https://www.youtube.com/watch?v=VOC3huqHrss
  • 9.
    Object Detection: History Ref.Sam Albanie, R-FCN: Region-based Fully Convolutional Networks, VGG Reading Group
  • 10.
    R-CNN: Pipeline Ref. RossGirshick et al., Rich feature hierarchies for accurate object detection and semantic segmentation, arXiv:1311.2524v5 [cs.CV] 22 Oct 2014
  • 11.
    R-CNN: Selective Search Ref.J. R. R. Uijlings et al., Selective Search for Object Recognition, IJCV 2013
  • 12.
    Object Detection: History Ref.Sam Albanie, R-FCN: Region-based Fully Convolutional Networks, VGG Reading Group
  • 13.
    Object Detection: Meta-architectures Ref.Jonathan Huang et al., Speed/accuracy trade-offs for modern convolutional object detectors, arXiv:1611.10012v3 [cs.CV] 25 Apr 2017
  • 14.
    Faster R-CNN: Architecture Ref.Shaoqing Ren et al., Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks, arXiv:1506.01497v3 [cs.CV] 6 Jan 2016
  • 15.
    Faster R-CNN: RegionProposal Network Ref. Shaoqing Ren et al., Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks, arXiv:1506.01497v3 [cs.CV] 6 Jan 2016
  • 16.
    Object Detection: Meta-architectures Ref.Jonathan Huang et al., Speed/accuracy trade-offs for modern convolutional object detectors, arXiv:1611.10012v3 [cs.CV] 25 Apr 2017
  • 17.
    R-FCN: Architecture Ref. JifengDai et al., R-FCN: Object Detection via Region-based Fully Convolutional Network, arXiv:1605.06409v2 [cs.CV] 21 Jun 2016
  • 18.
    R-FCN: Position-Sensitive ScoreMaps Ref. Jifeng Dai et al., R-FCN: Object Detection via Region-based Fully Convolutional Network, arXiv:1605.06409v2 [cs.CV] 21 Jun 2016
  • 19.
    Object Detection: Meta-architectures Ref.Jonathan Huang et al., Speed/accuracy trade-offs for modern convolutional object detectors, arXiv:1611.10012v3 [cs.CV] 25 Apr 2017
  • 20.
    YOLO: Architecture Ref. JosephRedmond et al., You Only Look Once: Unified, Real-Time Object Detection, arXiv:1506.02640v5 [cs.CV] 9 May 2016
  • 21.
    YOLO: Regression Model Ref.Joseph Redmond et al., You Only Look Once: Unified, Real-Time Object Detection, arXiv:1506.02640v5 [cs.CV] 9 May 2016
  • 22.
    Object Detection: Meta-architectures Ref.Jonathan Huang et al., Speed/accuracy trade-offs for modern convolutional object detectors, arXiv:1611.10012v3 [cs.CV] 25 Apr 2017
  • 23.
    Object Detection Benchmarks:Accuracy Ref. Jonathan Huang et al., Speed/accuracy trade-offs for modern convolutional object detectors, arXiv:1611.10012v3 [cs.CV] 25 Apr 2017
  • 24.
    Object Detection Benchmarks:GPU Time Ref. Jonathan Huang et al., Speed/accuracy trade-offs for modern convolutional object detectors, arXiv:1611.10012v3 [cs.CV] 25 Apr 2017
  • 25.
    Object Detection Benchmarks:Memory Usage Ref. Jonathan Huang et al., Speed/accuracy trade-offs for modern convolutional object detectors, arXiv:1611.10012v3 [cs.CV] 25 Apr 2017
  • 26.
    Object Detection Benchmarks:Accuracy vs. Time Ref. Jonathan Huang et al., Speed/accuracy trade-offs for modern convolutional object detectors, arXiv:1611.10012v3 [cs.CV] 25 Apr 2017
  • 27.
    Welcome to theworld where computer understands images.