The deep learning target detection framework requires GPU acceleration when performing image calculations, and requires the use of a hardware GPU graphics card. The connection between the target detection framework and the hardware GPU requires ① graphics card driver software; ② CUDA software to establish contact in turn. These two software can be downloaded directly from the NVIDIA official website. There are no very strict version requirements. According to our framework requirements, the CUDA version that can be installed by the graphics card driver software is greater than 11.1, and the CUDA version is 11.1.
1. Query the computer hardware GPU configuration (for querying the appropriate graphics card driver software)
① Find "This PC" on the desktop, or enter "This PC" in the folder interface, right-click and select "Manage" to enter the new interface
② Find "Device Manager" in the opening interface, follow the steps below, find the NVIDIA GPU graphics card, and get the graphics card version number. The following is NVIDIA GeForce GTX 1650
③Query the driver software according to the graphics card model, check the website:
https://www.nvidia.cn/Download/Find.aspx?lang=cn
④Correspondence between driver version and CUDA version:
⑤ According to the query configuration, the query results are as follows. It is recommended to download the driver version 470 and above.
This completes the installation of the driver software.
2. Install CUDA
① We clearly want CUDA11.1, so just find the corresponding version and install it.
Download address: https://developer.nvidia.com/cuda-downloads Open and enter the following interface
②Click to enter the following interface:
<1>2-Interface One:
<2>2-Interface Two
<3>2-Interface Three
<4>2-Interface Four
<5>2-Interface Five
③ Follow these steps to complete the download of CUDAToolKit. Click the exe during installation and it will be installed by default.
3. CuDNN download
In addition, there is CuDNN that needs to be downloaded and installed because it requires the deep learning library.
Download URL: https://developer.nvidia.com/rdp/cudnn-download
<1>3-Interface One
<2>3-Interface Two
<3 >3-Interface Three
<4>3-Interface Four
After downloading and decompressing, copy the three folders to the following installation directory of CUDA and overwrite them to complete the installation.
This completes the installation of all underlying software.