Cell segmentation using U-Net based networks

Table of Contents

• Project Pipeline
• GPU Support (Windows)
  • Tensorflow GPU
    • Tensorflow <= 2.10
    • Tensorflow > 2.10
  • Possible Errors
• GPU Support (Linux)

Project Pipeline

data > preprocessing > models > training > predictions

GPU Support (Windows)

Tensorflow GPU (Tensorflow <= 2.10)

You can follow from the official documentation here OR...

Guidance for Tensorflow Installation with CUDA, cudNN and GPU support: Youtube Video

Tested on Windows environment with Tensorflow 2.9, CUDA 11.2, cudnn 8.1

1. GO HERE FIRST to check cross compatibility

2. Download and install Microsoft Visual Studio

3. Download and install NVIDIA CUDA Toolkit

4. Download NVIDIA cuDNN

5. Extract cuDNN and transfer the files within

   • Copy folders bin, include, lib
   • Paste and replace to ...\NVIDIA GPU Computing Toolkit\CUDA\v11.2\
   • Search and open Edit the system environment variables
   • Go to Environment Variables
   • Double click on Path and add these full dir path:
     • ...\NVIDIA GPU Computing Toolkit\CUDA\v11.2\bin
     • ...\NVIDIA GPU Computing Toolkit\CUDA\v11.2\libnvvp

6. Install and create Anaconda Python environment (check compatible version)

conda create --name {any_name} python=={compatible_version}

7. Install Tensorflow (compatible with GPU support)

pip install tensorflow=={compatible_version}

Tensorflow GPU (Tensorflow > 2.10)

Please refer link

Possible Errors

Could not locate zlibwapi.dll. Please make sure it is in your library path

Copy of the missing zlib DLL in the NVIDIA Nsight directory:

C:\Program Files\NVIDIA Corporation\Nsight Systems 2022.4.2\host-windows-x64\zlib.dll

Copied and renamed it to:

C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.2\bin\zlibwapi.dll

GPU Support (Linux Ubuntu 22.04)

1. Reference 1
2. Reference 2