- Study of image processing a beginner depth direction I graduate. If wrong, I urge you to bear with me and predecessors pointing or two.
=============================================
Why install:
In fact, in-depth learning GPU regarded as a sign of the depth of entry to the study (personal understanding), before I use keras framework for the underlying CPU ran the most basic FCN32, FCN8, UNET, SEGNET, DEEPLABV-2-peer model, although with more time, but it will eventually be able to run up on your laptop and complete the training (every time I trained epoch = 50).
Until I used a paper model, the thesis segnet101 as basemodel, combined with mixed hollow convolution ASPP, and to dense upsampling to decode layer. . . Anyway, in my opinion it has been extremely complicated. However, the model used was mxnet framework, I did not use, and the code is not very good tune (probably I am cooking), so I grabbed the core code to keras modified into the framework of the codec layer, but in my when expectant ready for training, cpu too much, and this began my GPU do not stop the installation.
how to install:
Installation steps I divided into several parts:
1, computer-driven graphics view, and the corresponding mounting cuda cudnn
2, the configuration anacoda environment, install the corresponding version of the python, tensorflow-gpu and keras
In fact, seemingly simple, only two steps, but in reality was quite troublesome, I put a lot of times.
Now I focus on the next two steps, I try to have stepped pit and we talk about!
College freshman when I bought a computer, the graphics card is GTX960M, still usable
So first click on the desktop - "right click -" Open NVDIA interface, see Driver Version corresponding CUDA version, attach a correspondence table below:
(Red circle is covered 383.54)
White computer drive is 391.25, so the download CUDA9.0, Download refer https://blog.csdn.net/discoverer100/article/details/86696311
Then download the recent official website to download it seems not the line, so this time learned to find resources, such as looking for third-party help download Han (your product, you fine chemicals), I do not say it, after finding the resources to download the corresponding CUDNN, the corresponding windows10 figure below (the comments section need to leave your e-mail):
Then open the folder there are three files:
These three file folder to copy the contents of the folder corresponding to cuda, the contents will not be repeated: below shows the file path corresponding cuda
Then after the cuda files are configured, and set the environment variable path, set the environment variable method of self-examination about, do not understand the private letter or comment.
Before starting the second step, the drive is preferably updated (not updated to prevent too long a problem), the driving method of updating Open Device Manager - "point of the display configuration -" opening the properties - "Update drive means -" automatic search
Step 1 is complete it!
The second step below a blog Ha!
