【Introduction to Deep Learning & Computer Vision】Basic Knowledge Learning & Environment Construction & Advanced Tutorials & Practical Project Recommendations

1. Basic concepts

1. mAP (mean average precision): Target detection evaluation index, the average value of each category AP
2. Precesion: precision rate, that is, in the results returned after retrieval, the ratio of the truly correct number to the entire result
3. recall: The recall rate, that is, the ratio of the truly correct number in the retrieval results to the truly correct number in the entire data set (retrieved and unretrieved);
4. FN: False Negative, it is judged as a negative sample, but it is actually a positive sample;
5. FP: False Positive, it is judged as a positive sample, but it is actually a negative sample;
6. TN: True Negative, which is judged as a negative sample, is actually a negative sample;
7. TP: True Positive, judged to be a positive sample, in fact it is also a proof sample;
   * Memory method: Divide the abbreviation into two parts, the first letter (F, T) and the second letter (P, N). First of all, figure out the second letter, that is, what you think the attribution of the sample should be (Positive or Negative); the first letter is the evaluation of your judgment (False or True). Here maybe Chinese may be difficult to understand, so I want to describe it in English, which may be clearer: the second letter: What's your judgment about the sample?; the first letter: Is your judgment right(true) or not(false)?
   Then there is:
   precesion = TP/(TP+FP) That is, in the search results, there are all samples that you think should be positive (the second letter is all P), but there are the sums that you judged to be correct. Misjudged (the first letter has T, F).
   recall=TP/(TP+FN) That is, in the retrieval results, the samples that you judged to be positive are indeed positive, and those that are not judged as negative by you in the retrieval results but are actually positive (FN).
8. Explanation of terms in deep learning such as backbone, head, neck, etc.
9. NMS
10. Introduction to common model libraries:
    https://paddleclas.readthedocs.io/zh_CN/latest/models/index.html
11. IDE: VScode is recommended. Advantages: cross-platform, free, lightweight, and extensible; supports syntax highlighting, code auto-completion (IntelliSense), code refactoring, view definition functions, and built-in command line tools and Git version control system.
12. Anaconda: A set of software packages established to facilitate the use of Python for data science research, covering common Python libraries in the field of data science, and comes with a conda package management system specially designed to solve software environment dependencies. It mainly provides the functions of package management and environment management, which can easily solve the problems of coexistence and switching of multiple versions of python, as well as installation of various third-party packages. Anaconda uses the tool/command conda to manage packages and environments, and already includes Python and related supporting tools.
13. Cuda (can only be executed on devices with NVIDIA graphics cards): It is a parallel computing framework launched by NVIDIA that can only be used for its own GPU. The mainstream deep learning frameworks are also based on CUDA for GPU parallel acceleration.
14. Cudnn: An accelerated library for deep convolutional neural networks.
15. NVIDIA TensorRT: A high-performance neural network inference (Inference) engine for deploying deep learning applications in a production environment. Applications include image classification, segmentation, and object detection, etc., providing maximum inference throughput and efficiency.
16. Jupyter Notebook: A web-based application for interactive computing. It can be applied to the whole process of computing: development, documentation, running code and displaying results.

2. Environment deployment

1. Deep learning environment configuration Anaconda+Pycharm+CUDA+CUdnn+PyTorch+Tensorflow

2. Cuda installation tutorial

3. [Ubuntu 20.04 pytorch environment construction] Deep learning | Graphical installation process

4. The correct way to open - ANACONDA and jupyter installation

3. Advanced Tutorial

1. Automatic mixed precision training (based on paddlepaddle)
2. A complete explanation of the core basic knowledge of Yolov3&Yolov4&Yolov5&Yolox in the Yolo series
3. A complete explanation of the core basic knowledge of Yolov5 in the Yolo series
4. Migration Learning - Fine-tune
5. YOLOv5 achieves target detection (training your own data set for cat recognition)

4. Getting Started Project Recommendations

1. Super detailed yolov5 model training from scratch
2. Teach you to use YOLOV5 to train your own target detection model - mask detection - video tutorial
3. Deployment of hard hat detection YoloV3 model on Raspberry Pi (based on paddlepaddle)

5. Problem analysis and solution

1. Summary and analysis of low GPU utilization, low CPU utilization, and slow model training speed in deep learning PyTorch and TensorFlow

6. Reference learning materials

1. Hands-on deep learning PyTorch version (with GitHub source code, Jupyter Notebook learning)
   link:
   https://space.bilibili.com/1567748478/channel/seriesdetail?sid=358497
   https://zh.d2l.ai/index.html https://zh.d2l.ai/index.html
   https ://github.com/d2l-ai/d2l-zh

2. Wu Enda deep learning deeplearning.ai

3. PyTorch Deep Learning Quick Start Tutorial