# cv-sjy

**Repository Path**: hou-chenfeng/cv-sjy

## Basic Information

- **Project Name**: cv-sjy
- **Description**: 寒假深度学习项目，学习资料与论文代码
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2022-12-26
- **Last Updated**: 2022-12-29

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# 寒假深度学习项目



## 项目介绍

**项目日期：** 2023年1月8日-2月5日（春节假期调休）

**时间安排：** 直播课程一般在北京时间上午9-12点

**项目课时：** 60课时（每课时45分钟），包括核心课程、实践、科研写作等多个模块

**论文产出：** 在项目结束后学生将获得项目证书，同时可结合项目课题，产出一篇约5000词的科研论文（小组协作）

**课程大纲**:

<details markdown="1">
    <p>
    <br> 
    • Introduction: Neural Networks and Convolutional Processing 
	In this module, we are going to realize the perceptron and feed-forward networks, image filtering and processing and mathematics of convolutions and feature learning. Students will learn how to stacking  perceptrons to create networks, in order to learn it, training neural networks and convoluting processing is necessary. From convolutional layer to convolutional networks, we must know model local dependents on convolution, model strength dependents on hierarchy of features. Besides, we also will realize the image processing, which how to building block layer structure from 1D features to 2D processing.
    <br>
    • CNN Architectures
	In this module, we will learn the foundational CNN (convolutional neural network ) architecture. We are going to review LeNet: LeCun et al. 1998, AlexNet: Krizhevsky et al. Neurl PS 2012, GooLeNet/Inception: Szegedy et al. CVPR 2015, and VCGNet: Simomyan et al. ICLR 2015. During this period, we will realize the evolution of CNN, and find out the challenges in training CNNs. In addition to this, scaling CNNs also is the most part of the lecture and we will learn that from ResNet: He et al. CVPR 2016. The last part of this lecture is about the importance of data and how did datasets revolutionize computer vision.
    <br>
	• Sequential Image Processing
	In this module, we will discover the neural networks and sequential image processing. This module is composed of three parts, realizing Recurrent neural networks (RNN), how does RNNs and CNNs apply for 
vision, and applications in vision. We are going to realize the requirements for sequential processing, models for sequential processing, and recurrent neural networks for sequential modeling. We will research how to use multimodal CNN + RNN architecture for captioning and how to use video data predict action.
    <br>
	• Generative Image Modeling
	In this course, we will discover what is generative modeling and why generative modeling. This part includes density estimation, structured prediction and synthesis. Besides, realizing taxonomy of generative models also is the most part. We will find out that variational autoencoders will lead to reconstruction results. We also will discuss what is Generative adversarial networks (GANs) , compare traditional generation and StyleGAN, and find latest results about StyleGAN2.
    <br>
	• Neural Graphics and Rendering
	In this module, we are going to discuss classical computer graphics, neural scene representations and implicit neural rendering. We will find out that rendering is a complex process and its differentiation is not uniquely defined, which prevents straightforward integration into neural networks. Differentiable rendering constitutes a family of techniques that tackle such an integration for end-to-end optimization by obtaining useful gradients of the rendering process.
    <br>
	• Neural Vision Applications 
	In this course, we will present some neural vision applications, including object detection, semantic segmentation, and selfsupervised vision. The traditional object detection method uses the sliding window method to detect image region by region, but with the wide application of deep convolutional neural networks, Grishick et al. proposed R-CNN target detection framework. We will go deep on R-CNN. Semantic segmentation often requires the extraction of features and representations, which can derive meaningful correlation of the input image. This lecture will focus on the FCNN module. 
    <br>
	• Interpretability and Uncertainty in Computer Vision
	In this module, we are going to learn about uncertainty estimation in computer vision. We hope that the model can bring uncertainty and help people who use the model to make better decisions. Generally, there are two types of uncertainty, aleatoric uncertainty and epistemic uncertainty. We will research these two kinds of uncertainty in depth and realize the application of model uncertainty in life. 
    <br>
	• Computer Vision Progress and Perspectives 
	In this course, we will research the progress and perspectives of computer vision in several areas, such as facial recognition, performance advances in generative models, AI-generated art, computer vision in robotics, autonomous vehicle training and control, and computer vision in safety-critical application. We will get better idea from specific case of image classification. Besides, we also will learn the bias in computer vision and how to debiasing it.
        </p></details>











## 预习：python与深度学习框架



TODO：

- 安装 anaconda pycharm
- jupyter语法基础、anaconda使用
  - jupyterhttps://zhuanlan.zhihu.com/p/63186778
  - anacondahttps://blog.csdn.net/Alexa_/article/details/123966189
  - pip https://zhuanlan.zhihu.com/p/526821965 anaconda中有conda替代pip
- python语法 和numpy学习
  - 预习包中python基础学习的文件，点击使用jupyter运行
  - 参考 马士兵python教程 https://www.bilibili.com/video/BV1Dd4y1S7gP/?spm_id_from=333.337.search-card.all.click&vd_source=a3f7279ad89529aa08a8a0fcf24cea6d
  - 参考 浙大python慕课 https://www.icourse163.org/learn/ZJU-1206456840?tid=1466879462#/learn/content
  - cs231 的numpy教程（推荐）和资料中的pdf一样，可以colab运行，语法同jupyter https://cs231n.github.io/python-numpy-tutorial/
  -  华为numpy教程 pytorch框架 https://education.huaweicloud.com/courses/course-v1:HuaweiX+CBUCNXE081+Self-paced/about


​    

深度学习入门参考：

- 华为 深度学习 基本概念 https://education.huaweicloud.com/courses/course-v1:HuaweiX+CBUCNXE088+Self-paced/about
- 达尔闻 计算机视觉入门 opencv入门（强推）https://www.bilibili.com/video/BV1Ui4y137pW/?spm_id_from=333.788.video.desc.click&vd_source=a3f7279ad89529aa08a8a0fcf24cea6d
- 小土堆pytorch教程（强推）https://www.bilibili.com/video/BV1hE411t7RN/?spm_id_from=333.337.search-card.all.click
- git 语法https://blog.csdn.net/youzhouliu/article/details/78952453

## 文献调研 

### 建筑*深度学习

<<<<<<< HEAD
黑客马拉松 建筑+ai：一键排砖、基于功能拓扑关系的智能平面图生成平台

=======
黑客马拉松 建筑+ai
　　一键排砖、基于功能拓扑关系的智能平面图生成平台
>>>>>>> b8252cf532a18afd7acae33b0541bffce5fd3b20
 http://www.chuangyisai.com/announce/jxhjys/jxjzsj/2403.html

建筑+ai的独角兽公司 小库科技

http://www.archcollege.com/archcollege/2021/12/50283.html

#### 文献：

- 设计相关：生成类模型，根据文本或图像 生成符合要求的 图像、视频、三维模型
- 三维重建相关：nerf街景重建  
- 待续。。。



平面图生成、

项目论文