tutorial.md

mmv_im2im Docker Deployment

Installation

Prerequisite:

• You need to download the docker for your operating system, see the tutorial here.
• To utilize GPU, it is also required to install nvidia-docker.
• on MacOS, we recommend to allocate at least 8GB memory and 4GB swap for docker to run our example.

1. Arm64(Apple M1/2)

Firstly, pull our image from the dockerhub:

docker pull mmvlab/mmv_im2im:v0.4.0_arm64

Then create and run a container:

# make sure you are in the root dir of mmv_im2im package
bash docker/arm64/run_container.sh v0.4.0_arm64

2. Amd64(Intel/AMD CPU)

Firstly, pull our image from the dockerhub:

docker pull mmvlab/mmv_im2im:v0.4.0_amd64

Then create and run a container:

# make sure you are in the root dir of mmv_im2im package
bash docker/amd64/run_container.sh v0.4.0_amd64

3. CUDA(Nvidia GPU)

Firstly, pull our image from the dockerhub:

docker pull mmvlab/mmv_im2im:v0.4.0_amd64_cu113

Then create and run a container:

# make sure you are in the root dir of mmv_im2im package
bash docker/cuda/run_container.sh v0.4.0_amd64_cu113

Simple tutorial: labelfree 2d task

We illustrate the usability of our package through a simple labelfree 2d task.

• To download the example data, please refer to this notebook. Please make sure the data is in the right path.
• We recommend you to run the docker using vscode with docker plugin.
• To run the code:
  • for training:

  run_im2im --config_path 'paper_configs/labelfree_2d_FCN_train.yaml'\
    --data.data_path 'data/labelfree2D/train'\
    --trainer.params "{'max_epochs':10,'accelerator':'auto'}"\
    --data.dataloader.train.dataloader_params "{'batch_size':1,'num_workers':1}"

  • for testing:

  run_im2im --config_path 'paper_configs/labelfree_2d_FCN_inference.yaml'\
    --data.inference_input.dir 'data/labelfree2D/test'\
    --data.inference_output.path 'data/labelfree2D/pred'\
    --model.checkpoint 'lightning_logs/version_0/checkpoints/best.ckpt'