Fresh Install of a Mask-RCNN Environment on the Google Cloud Platform

26 August 2020 - 2 mins read time

Fresh MRCNN Installation Instructions

ssh into the gcloud instance
Install the conda environment. Here are the instructions. Only follow till step 8, as making the conda environment will require a specific instruction.
Download the Matterport implementation of the Mask-RCNN library
```
 git clone https://github.com/matterport/Mask_RCNN.git
```
Change directory to Mask_RCNN
```
 cd Mask_RCNN
```

Download the working-gpu.txt file

 wget https://github.com/matterport/Mask_RCNN/files/3870791/working-gpu.txt

Create the conda environment using the file flag

 conda create --name <env-name> --file working-gpu.txt

Activate the conda environment
```
 conda activate <env-name>
```
Run the setup script
```
 python3 setup.py install
```

Check for GPU details.

 nvidia-smi

(If the above fails)

 sudo add-apt-repository ppa:graphics-drivers/ppa
 sudo apt-get update
 sudo apt-get install nvidia-driver-418

(Check again)

 nvidia-smi

Make a data directory for gcsfuse to connect your training data bucket to
```
 cd ~ && mkdir data
```

Notes

If you’re having ipython/jupyter notebook issues since it is using the wrong python kernel, read here
If you’re having version compatibilty issues for keras, tensorflow, etc.. follow this matrix:

Keras	Tensorflow	Tensorflow-GPU	cuDNN	cuda-toolkit	nvidia-drivers
2.0.8	1.14	1.14	7.4	10.1	418.39

If the python script using the GPU was terminated but the GPU still hasn’t cleared its memory check your GPU processes

  sudo fuser -v /dev/nvidia*

The output will look something like

  USER        PID  ACCESS COMMAND
  /dev/nvidia0:        root       1256  F...m  Xorg
  username   2057  F...m  nvidia-persis...
  username   20699 F...m  python

Terminate the python process

  sudo kill -9 PID