Shan Ye
data-driven geoscientist, software engineer, AI4S researcher
Contact
Email: yeshantulsa@hotmail.com
Chinese webpage: 中文网页
Social Media
计算机视觉 Computer Vision
知识图谱 Knowledge graph of the class
课件 Slides
| 节次/Lec | 主要内容/Content | |
|---|---|---|
| 1. 概述 (intro) | 课程概况、行业概况、学科关系、哲学基础、发展历史 (overview, related disciplines, fundamentals, history) | |
| 2. 硬件 (hardware) | 机器视觉系统的构成、光源、镜头、相机、感光器件 (light source, lens, camera, photosensitive device, etc.) | |
| 3. 图像增强1 (image preprocess 1) | 图像的构成及常见格式、色彩空间、锐化和降噪、滤波器 (data structure of images, color space, noise reduction, filtering, etc.) | |
| 4. 图像增强2 (image preprocess 2) | 卷积和滤波、边缘检测、图像金字塔、FFT及频域操作 (convolution and filtering, edge detection, image pyramid, Fourier transform, frequency domain operations, etc.) | |
| 5. 图像分类任务1 (image classification 1) | 使用场景、任务类别、难点和挑战 (classification tasks, challenges, etc.) | |
| 6. 图像分类任务2 (image classification 2) | 图像的向量表示、图像的特征表示、机器学习流程 (vector representation of images, features of images, review on machine learning basics) | |
| 7. 线性分类器1 (linear classifier 1) | 线性代数表达、决策边界、合叶损失 (linear algebra basics review, decision boundary, Hinge loss, etc.) | |
| 8. 线性分类器2 (linear classifier 1) | 超参数、正则项、梯度计算 (hyperparameters, regularization, gradient, etc.) | |
| 9. 模型的训练 (training a model) | 随机梯度下降、小批量梯度下降、交叉验证 (stochastic gradient descent, mini-batch gradient descent, cross validation, etc.) | |
| 10. 全连接神经网络1 (fully-connected neural networks 1) | 网络结构、激活函数、交叉熵损失和KL散度、信号传播、梯度回传 (ANN structure, activation function, cross entropy loss and KL divergence, signal propagation, backpropagation, etc.) | |
| 11. 全连接神经网络2 (fully-connected neural networks 1) | 梯度消失和梯度爆炸、动量法、自适应梯度、自适应矩估计、权值初始化、批归一化、随机失活、调参 (gradient disappearance and gradient explosion, momentum method, AdaGrad, Adam, weight initialization, batch normalization, dropout, parameter tuning, etc.) | |
| 12. 卷积神经网络1 (convolutional neural network 1) | CNN总体架构、纹理基元、卷积核组 (CNN architecture, image texture, convolution kernel group) | |
| 13. 卷积神经网络2 (convolutional neural network 2) | 卷积层的堆叠、池化层 (stacking of convolutional layers, pooling) | |
| 14. 经典构架1 (classic CNNs 1) | LeNet, AlexNet | |
| 15. 经典构架2 (classic CNNs 2) | ZF Net, VGG, GoogLeNet, ResNet | |
| 16. 前沿进展 (recent advances) | vision transformer, diffusion models, GANs |
每个节次90分钟 (90 mins per lecture)
Lab 课作业
| 节次/Lec | 主要内容/Contents | |
|---|---|---|
| Lab 1 | 环境配置、Python练习、图像的基本操作 (working environment configuration, Python basics, basic image operations) | |
| Lab 2 | 卷积和滤波、OpenCV的基本操作 (convolution, filtering, basics of OpenCV) | |
| Lab 3 | 图像金字塔 (image pyramid) | |
| Lab 4 | 傅里叶转换和频率域操作 (Fourier transforms and frequency domain operations) | |
| Lab 5 | 线性分类器的实现 (coding up a linear classifier) | |
| Lab 6 | 计算图的OOP实现、梯度计算 (object-oriented programming for a computation graph, gradient calculation) | |
| Lab 7 | ResNet和迁移学习 (coding up a ResNet, transfer learning) | |
| Lab 8 | 风格迁移 (style transfering) |
说明 Notes
本课程部分内容参考了以下课程 (This course is partially derived from the following classes)：