Learning Record 1 of Computer Vision and Machine Learning (environment configuration and installation + note sharing)

1. Precautions for Pytorch and CUDA installation

        ①According to the reference [3], check the GPU driver version used by the computer through the NVIDA control panel;

        ② According to reference [2] and computer GPU driver version, check the corresponding CUDA version of the graphics card driver;

        ③According to the reference [3], download the corresponding versions of CUDA and cuDNN respectively.

       Note : The computer GPU driver version may not find the corresponding download version of CUDA, but CUDA follows the principle of backward compatibility. Therefore, we can choose the latest old version of the corresponding version of the GPU driver to download CUDA; Jupyter's Pytorch environment installation can refer to [1].

---

 (1) GPU

        A Graphics Processing Unit (GPU) is dedicated to performing the intensive calculations required to render images, video, and animations on a computer. It has a large number of cores, but the calculation speed of each core is much slower, which is very suitable for simple parallel tasks.

(2) CUDA

        CUDA (Computer Unified Device Architecture) is a computing platform launched by graphics card manufacturer NVIDIA. CUDA TM is a general-purpose parallel computing architecture introduced by NVIDIA, which enables GPUs to solve complex computing problems. It includes CUDA instruction set architecture (ISA) and parallel computing engines inside GPUs.

(3) cuDNN

        The NVIDIA CUDA Deep Neural Network Library (cuDNN) is a GPU-accelerated library of primitives for deep neural networks. cuDNN provides highly optimized implementations of standard routines, such as forward and backward convolution, pooling, normalization, and activation layers, among others.

        Deep learning researchers and framework developers around the world rely on cuDNN for high-performance GPU acceleration. It allows them to focus on training neural networks and developing software applications, rather than spending time on low-level GPU performance tuning. cuDNN accelerates widely used deep learning frameworks including Caffe2, Chainer, Keras, MATLAB, MxNet, Pytorch, and TensorFlow.

---

2. Some practical considerations

(1) When designing your own network, you need to pay attention to the __init__ function and the forward function should have the same process structure. Conv2d, ReLU, BatchNorm2d, MaxPool2d and other functions in the __init__ function cannot be instantiated only once because the input parameters are consistent. , functions with the same parameter used multiple times must be instantiated with different names;

(2) In practice, the reasonable division ratio of the training set, verification set and test set of the sample set is usually 6:1:3, and the division ratio should be based on individual sample categories.

3. Computer vision and machine learning study notes (entry system)

        Entry-Level Notes for Computer Vision and Machine Learning-Python Documentation Resources-CSDN Download

References:

[1] Install pytorch in win10 anaconda3 environment and use it in jupyter

[2] The cuda version corresponding to the graphics card driver_kyle-fang's blog-CSDN blog_cuda version and graphics card version

[3] pycharm configures pytorch_huanglianghuang's blog-CSDN blog_pycharm pytorch

[4] Pytorch neural network programming teaching for beginners_哔哩哔哩_bilibili

[5]deeplizard - Resources

[6] [pytorch Chinese documentation] torch.nn - pytorch Chinese Network

[7] 5-Automatic derivative mechanism_哔哩哔哩_bilibili (Tang Yudi course)

[8] Training model building_哔哩哔哩_bilibili

[9] pytorch commonly used loss function finishing article (1)_L1_Zhang's blog-CSDN blog_pytorch error function

[10] Intensive reading of ResNet papers paragraph by paragraph [Intensive reading of papers] - Zhihu (ResNet explanation)

[11]  Deep Learning and CV Tutorial (2) | Basics of Image Classification and Machine Learning - ShowMeAI - 博客园