Get best of the week

Setting up the neural network environment Mask R-CNN

Good day, as part of the study of neural networks, many encounter difficulties in setting up the environment. To this end, I decided to write an article in order to help thirsty beginners.

As part of his task, he used the architecture of Mask R-CNN .

Links to all distributions will be offered at the end of the post.

So, let's start, in order to be able to use the functionality of neural networks, like any software, it must be installed and configured. This paper presents the following set of actions:

  1. Remove all NVIDIA drivers / software and remove all NVIDIA files from program files (also x86). It is necessary so that the old drivers do not affect the operation of newly installed components.
  2. Install Visual Studio [1]. During the installation process, together with the IDE, it supplies the libraries necessary for further work, including Visual C ++. The minimum version of the program is 2015.
  3. Install CUDA [2] - a software and hardware architecture for parallel computing, which can significantly increase computing performance through the use of Nvidia GPUs. The minimum version is 9.0.
  4. Download cuDNN [3] is a library for working with deploying neural networks with support for graphic accelerators. The version must match the CUDA version.
  5. Copy the contents of the downloaded and extracted cuDNN folder to the appropriate folders in C: \ Program Files \ NVIDIA GPU Computing Toolkit \ CUDA \ Vx.x \.
  6. Go to the environment variables of your system and edit the path and add the following two directories to your path:

    • C: \ Program Files \ NVIDIA GPU Computing Toolkit \ CUDA \ v9.0 \ bin
    • C: \ Program Files \ NVIDIA GPU Computing Toolkit \ CUDA \ v9.0 \ libnvvp
  7. Download and install Python [4] - a high-level general-purpose programming language aimed at improving developer productivity and code readability. The minimum version is 3.5.
  8. Anaconda [5] – Python R, , . Anaconda to PATH , , IDE Windows CMD. Python.
  9. Windows CMD , :
    conda create -n [env_name] python = [python_version]
  10. , ,
    activate [env_name]
  11. tensorflow ( 1.13.1) keras ( 2.3), , :

    pip install --ignore-connected --upgrade tenorflow-gpu
	pip install keras
	pip install cv2
	pip install numpy ()
        pip install pandas ()
	pip install matplotlib ()
	pip install pillow ()

Installation is done through the package management system - pip, which comes with the IDE. Commands are entered in Windows CMD.

After all the steps described, you can begin to build a model and train the network.

Promised distribution links:

  1. visualstudio.microsoft.com/downloads
  2. developer.nvidia.com/cuda-90-download-archive
  3. developer.nvidia.com/rdp/cudnn-download
  4. www.python.org/downloads
  5. www.anaconda.com/products/individual

More articles:


All Articles