강의 복습 내용

[DAY 13] Convolutional Neural Networks

[DLBasic] CNN - Convolution은 무엇인가?

• Convolutional Neural Networks
  • CNN consists of convolution layer, pooling layer, and fully connected layer.
    • Convolution and pooling layers: feature extraction
    • Fully connected layer: decision making (e.g., classification)
• Stride
  
• Padding
  
• Stride & Padding
  

[DLBasic] Modern CNN - 1x1 convolution의 중요성

• ILSVRC(ImageNet Large-Scale Visual Recognition Challenge)

  • Classification / Detection / Localization / Segmentation
  • 1,000 different categories
  • Over 1 million images
  • Training set: 456,567 images

• AlexNet

  • Key ideas
    • Rectified Linear Unit (ReLU) activation
    • GPI implementation (2 GPUs)
    • Local response normalization, Overlapping pooling
    • Data augmentation
    • Dropout
  • ReLU Activation
    • Preserves properties of linear models
    • Easy to optimize with gradient descent
    • Good generalization
    • Overcome the vanishing gradient problem

• VGGNet

  • Increasing depth with 3 × 3 convolution filters (with stride 1)
  • 1x1 convolution for fully connected layers
  • Dropout (p=0.5)
  • VGG16, VGG19
  • Why 3 × 3 convolution?

• GoogLeNet

  • GoogLeNet won the ILSVRC at 2014
    • It combined network-in-network (NiN) with inception blocks.
  • Inception blocks
    • What are the benefits of the inception block?
      • Reduce the number of parameter.
    • How?
      • Recall how the number of parameters is computed.
      • 1x1 convolution can be seen as channel-wise dimension reduction.
  • Benefit of 1x1 convolution
    • 1x1 convolution enables about 30% reduce of the number of parameters!

• Quiz

  • Which CNN architecture has the least number of parameters?
    1. AlexNet (8-layers) (60M)
    2. VGGNet (19-layers) (110M)
    3. GoogLeNet (22-layers) (4M)
  • The answer is GoogLeNet.

• ResNet

  • Deeper neural networks are hard to train.
    • Overfitting is usually caused by an excessive number of parameters.
    • But, not in this case.
  • Add an identity map (skip connection)
  • Add an identity map after nonlinear activations:
  • Batch normalization after convolutions:
  • Performance increases while parameter size decreases.

• DenseNet

  • DenseNet uses concatenation instead of addition.
  • Dense Block
    • Each layer concatenates the feature maps of all preceding layers.
    • The number of channels increases geometrically.
  • Transition Block
    • BatchNorm -> 1x1 Conv -> 2x2 AvgPooling
    • Dimension reduction

• Summary

  • VGG: repeated 3x3 blocks
  • GoogLeNet: 1x1 convolution
  • ResNet: skip-connection
  • DenseNet: concatenation

[DLBasic] Computer Vision Applications

• Fully Convolutional Network
  • This is how an ordinary CNN looks like.
    
  • This is a fully convolutional network.
    
  • Transforming fully connected layers into convolution layers enables a classification net to output a heap map.
  • While FCN can run with inputs of any size, the output dimensions are typically reduced by subsampling.
  • So we need a way to connect the coarse output to the dense pixels.
• Deconvolution (conv transpose)
  
• Detection
  • R-CNN
    1. takes an input image,
    2. extracts around 2,000 region proposals (using Selective search),
    3. compute features for each proposal (using AlexNet), and then
    4. classifies with linear SVMs.
  • SPPNet
    • In R-CNN, the number of crop/warp is usually over 2,000 meaning that CNN must run more than 2,000 times (59s/image on CPU).
    • However, in SPPNet, CNN runs once.
  • Fast R-CNN
    1. Takes an input and a set of bounding boxes.
    2. Generated convolutional feature map
    3. For each region, get a fixed length feature from ROI pooling
    4. Two outputs: class and bounding-box regressor.
    • Faster R-CNN = Region Proposal Network + Fast R-CNN
  • YOLO
    • YOLO (v1) is an extremely fast object detection algorithm.
      • baseline: 45fps / smaller version: 155fps
    • It simultaneously predicts multiple bounding boxes and class probabilities.
      • No explicit bounding box sampling (compared with Faster R-CNN)
    • Given an image, YOLO divides it into SxS grid.
      • If the center of an object falls into the grid cell, that grid cell is responsible for detection.
    • Each cell predicts B bounding boxes (B=5).
      • Each bounding box predicts
        • box refinement (x / y / w / h)
        • confidence (of objectness)
    • Each cell predicts C class probabilities.
    • In total, it becomes a tensor with SxSx(B*5+C) size.
      • SxS: Number of cells of the grid
      • B*5: B bounding boxes with offsets (x,y,w,h) and confidence
      • C: Number of classes

[DLBasic] CNN - 강아지 종류 분류하기

• Self-study guide
  • Dog breed 데이터셋의 다운로드부터 Dataloader 생성까지의 전 과정을 기존의 CNN 모델 파일(py) 파일 수정하여 작성해 볼 것

[DLBasic] CNN - 나만의 데이터셋 만들기

• Self-study guide
  • 팀별로 수집할 데이터 주제를 선정한다.
  • 구글을 통해 관련된 데이터를 다운로드 받는다.
  • 같은 class의 데이터를 폴더별로 모은다.
  • 해당 데이터중 관련이 없는 데이터를 삭제하거나 새로운 분류를 만들어 따로 모은다.
  • CNN 모델을 만들어 학습한다.

Further Question

수업에서 다룬 modern CNN network의 일부는, Pytorch 라이브러리 내에서 pre-trained 모델로 지원합니다. pytorch를 통해 어떻게 불러올 수 있을까요?

피어 세션 정리

강의 리뷰 및 Q&A

• [DLBasic] CNN - Convolution은 무엇인가?
• [DLBasic] Modern CNN - 1x1 convolution의 중요성
• [DLBasic] Computer Vision Applications
• [DLBasic] CNN - 강아지 종류 분류하기
• [DLBasic] CNN - 나만의 데이터셋 만들기

과제 진행 상황 정리 & 과제 결과물에 대한 정리

[DLBasic] CNN Assignment

CNN(Convolutional Neural Network)에 대한 것으로, 어렵지 않게 해결했습니다.

총평

이론 공부도 재미있지만, 역시 저는 어떤 목적을 가지고 문제를 해결할 때가 가장 즐겁고 많이 배우는 것 같습니다.
그래서 오늘 강의에서 언급된 Self-study를 위한 실습이 너무 반가웠습니다.
해당 실습을 통해서, PyTorch를 가지고 놀다 보면 금방 익숙해질 것 같습니다.
물론, 데이터 수집부터 전처리, 모델링까지 해야 할 일이 더 늘었지만, 다음 주를 이용하면 어떻게든 해낼 수 있다고 생각합니다.

오늘보다 더 성장한 내일의 저를 기대하며, 내일 뵙도록 하겠습니다.

읽어주셔서 감사합니다!