Base-Pytorch-Trainer
Project structure
.
├── configs: Put configure files
├── dataset: Put dataset files
├── submissions: Put submission files
├── src
│ └── main
│ ├── data: Put data-processing code
│ ├── engine: Put trainer, predicter, and splitter code
│ ├── model: Put model code
│ ├── options: Put argparse code
│ ├── predict.py: Predict entry
│ ├── split_dataset.py: Split dataset entry
│ └── train.py: Train entry
├── do_predict.sh
├── do_split_dataset.sh
├── do_tensorboard.sh
└── do_train.sh
Tips
If you are using pycharm, set src/main as sources root.
How to use
1, Install requirements
pip install tensorboard
conda install pytorch torchvision torchaudio cudatoolkit=10.2 -c pytorch pyyaml
2, Put your dataset to ./dataset
A generic dataset folder where the images are arranged in this way by default:
root/train_images/label1/xxx.png
root/train_images/label1/xxx.png
...
root/train_images/label2/xxx.png
root/train_images/label2/xxx.png
...
...
root/test_images/xxy.png
root/test_images/xxy.png
...
3, Set your parameter
You can set parameter on config/xxx.yaml which will be set as default parameter.
Then, you can override parameter on shell script.
4, Split dataset
Running Script ./do_split_dataset.sh
5, Create model
Create your model on src/main/model
6, Train
Running Script ./do_train.sh
7, Predict
Running Script ./do_predict.sh
Every row of submission file should already have ID
