CNN for VR A4：Google Cloud with GPUs Tutorial

目录

• 1 Image Classification Introduction
  • 1.1 Nearest Neighbor Classifier
    • 1.1.1 k-Nearest Neighbor
  • 1.2 Train/Val/Test Split
  • 1.3 Pros/Cons of Nearest Neighbor
• 2 Summary
  • 2.1
  • 2.2 Applying kNN in practice
• 6 参考资料

---

1 Image Classification Introduction

Image Classification：给出1幅图片，在固定的标签(Label)中给出图片中对应物体的概率分布。这个任务定义看起来简单，实现起来却不容易。而且很多Computer Vision task最后都可化约为Image Classification问题。

Image Classification有如下几个主要难点：

1. 视角变化(Viewpoint variation)

2. 大小变化(Scale variation)

3. 变形(Deformation)

4. 遮盖(Occlusion)

5. 光照变化(Illumination Conditions)

6. 背景杂波(Background Clutter)

7. 类内变化(Intra-class Variation)

1.1 Nearest Neighbor Classifier

1.1.1 k-Nearest Neighbor

1.2 Train/Val/Test Split

1.3 Pros/Cons of Nearest Neighbor

2 Summary

2.1

2.2 Applying kNN in practice

repeat until convergence：

$$\theta_0 := \theta_0 - \alpha \frac 1 m \sum_{i=1}^m[h_\theta(x^{(i)})-y^{(i)}]$$ $$\theta_1 := \theta_1 - \alpha \frac 1 m \sum_{i=1}^m \lbrace[h_\theta(x^{(i)})-y^{(i)}]x^{(i)}\rbrace$$

6 参考资料

• “CS231n: Convolutional Neural Networks for Visual Recognition”;
• “CS231n Course Video 2016 Winter”;
• “Deep Learning”;