Compute Engine > VM instances > Create instance. Region & Zone: same zone as your GPU quota request Machine type: 2 vCPUs and 7.5 GB memory is a good starting point. If training 2000+ images for each face, use 16+ GB RAM. The cost will increase. Click Customize> Number of GPUs: 1 GPU Type: NVidiaTesla P100 GPU (or K80 if that's what you got approved) Boot disk: Windows Server 2016 Datacenter You can choose a Regular Persistent Disk or SSD (more expensive). Firewall: Allow HTTP; Allow HTTPS If you requested a Preemptible GPU: Click Management, security, disks, etc > Preemptibility: On Create! Once your instance is created, Google will start charging your free trial.
Install Microsoft RDP (OS X App Store link) There is also an RDP Chrome extension, but it can be unreliable. Navigate to your instance from the instance panel. Click "Set Windows password" COPY the password somewhere SAFE! Don't lose it! 👹 Click on the ▾ button next to "RDP" > > Download RDP file. Open it & log on with your password. You're in! 🙃
Download Chrome immediately because Internet Explorer is annoying. You might need to edit the Internet Explorer security settings to do this. Now, we will need: NVIDIA Graphics Driver (win64bit K80) (win64bit P100) Visual Studio Visual C++ Redistributable 64 bit CUDA 9.2 (express install) Check PATH variable paths: Search > "environment">edit environment variables Environment variables...>Path>Edit... Paths should be:C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.2\binC:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.2\libnvvpOtherwise, manually find & select installation dir \CUDA\v9.2\bin FakeApp > unzip to C:\ FFmpegAdd the bin directory to your PATH environmentC:\ffmpeg-3.4.1-win64-static\bin Daum PotPlayer 64 bit Lightning Image Resizer
Microsoft Visual Studio Redistributable 2015 CUDA 9.0 (not 9.1) ❊ Custom installation; ignore C++ warnings, don't install drivers cuDNN 7.05 ❊ Unzip the cuDNN files. ❊ You need to manually copy 3 files from cuDNN to the corresponding folders in your CUDA installation, which is usually at C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.0 ❊ Copy cuda\bin\cudnn64_7.dll to C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.0\bin ❊ Copy cuda\include\cudnn.h to C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.0\include ❊ Copy cuda\lib\x64\cudnn.lib to C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.0\lib\x64 NVIDIA graphics card drivers Download and run OpenFaceSwap ❊ Install in a directory that isn't "Program Files" (possible user permission errors). ❊ The first time you run it, it may take a few minutes for anything to happen. You can check Task Manager to make sure it's active.
DeepDenial Workshop Intro + Group ExercisesTraining Data Extraction