Project Organization

``` ├── .github/ <- Settings for GitHub. │ ├── data/ <- Datasets and image processing code (data Organization is documented in "data/README.md") │ ├── environments/ <- Provision depends on environments. │ ├── models/ <- Pretrained and serialized models. │ ├── notebooks/ <- No use │ ├── outputs/ <- Outputs │ ├── src/ <- Source code. │ ├── tests/ <- No use │ ├── .flake8 ├── .dockerignore ├── .gitignore ├── LICENSE <- Ascender LICENSE ├── Makefile <- No use ├── poetry.lock <- Lock file. DON'T edit this file manually. ├── poetry.toml <- Setting file for Poetry. ├── pyproject.toml <- Setting file for Project. (Poetry, Black, isort, Mypy) ├── main.py <- main code (Train and Test) ├── prediction.py <- Predictin code (Create Prediction images 512x512 pix) └── README.md <- The top-level README for developers.

```

Using template repository

• https://github.com/cvpaperchallenge/Ascender　\ If there are any issues with the setup, a solution may be found at the above URL

PC setting

• OS:Ubuntu20.04LTS
• CPU:Intel® Core™ i9-10900 CPU @ 2.80GHz × 20
• GPU: NVIDIA GeForce RTX 3090
• Memory: 64GB

SetUp

Create accounts

• install docker
• create Wandb account from here
  • copy user name and API key
• create GitHub account from here
• create figshare account from here

Download code from github

shell UserNmae@yourPC: $ git clone git@github.com:tohsato-lab/S-loss.git

Create docker container and install python3 virtual environment

shell UserNmae@yourPC:S-loss $ cd environment/gpu UserNmae@yourPC:S-loss/environment/gpu $ docker compose up -d UserNmae@yourPC:S-loss/environment/gpu $ docker compose exec core bash challenger@hoge:~/ascender $ poetry install

Downloading datasets

Please check the folder structure from S-loss/data/README.md . 1. Download labels ("NeuroGT labeldata") and datatables ("contours") from here 2. Download NeuroGT Images and resize with downloadNeurogt.sh ```shell UserNmae@yourPC:S-loss $ cd environment/gpu UserNmae@yourPC:S-loss/environment/gpu $ docker compose exec core bash challenger@hoge:~/ascender $ cd data challenger@hoge:~/ascender/data $ bash downloadNeurogt.sh ```

Wandb setting

• wandb setting in docker container shell challenger@hoge:~/ascender $ poetry run wandb init Let's setup this directory for W&B! wandb: Logging into wandb.ai. (Learn how to deploy a W&B server locally: https://wandb.me/wandb-server) wandb: You can find your API key in your browser here: https://wandb.ai/authorize wandb: Paste an API key from your profile and hit enter, or press ctrl+c to quit: {API key} wandb: Which team should we use? wandb: (1) {Youre User Name} wandb: (2) Manual entry wandb: Enter your choice: 1 wandb: Which project should we use? wandb: (1) Create New wandb: Enter your choice: 1 wandb: You chose 'Create New' Enter a name for your new project: {Project Name} This directory is configured! Next, track a run: challenger@hoge:~/ascender $ poetry run wandb init

Running

Running Train and Test

1. Edit main.py python {main.py} if __name__ == '__main__': CFG = get_args() if CFG.wandb_logging: run = wandb.init( project="{Project Name}", # ← here config={k: v for k, v in dict(vars(CFG)).items() if '__' not in k}, name=f"{CFG.Name}", entity="{User Name}" # ← here ) wandb.save(os.path.join(CFG.Homepath, 'src', 'train_core.py')) main_loop(CFG)
2. Running main.py (Train and Test) shell challenger@hoge:~/ascender $ poetry run python3 main.py

Running Prediction and Create BD5 dataset

1. download model weight (hogehoge.pth) from wandb and save in " models/ " folder
2. Edit prediction.py python model = load_model.load_deplaboV3plus(configs=configs) model_name = f'model_weight_name' # ←here model_path = os.path.join(configs.Homepath, "models", model_name + '.pth') model.load_state_dict(torch.load(model_path))
3. Running prediction.py shell challenger@hoge:~/ascender $ poetry run python3 predinction.py
4. Resizing and Create BD5 shell challenger@hoge:~/ascender $ cd data challenger@hoge:~/ascender/data $ poetry run python3 resize.py challenger@hoge:~/ascender/data $ poetry run python3 create_h5_split_class_id.py