This repository is a system for training image classification models in pytorch using MLOps techniques with easy customization for problems.
To train the models on your data follow these steps:
- Extract the dataset into the data folder. Obs: the data must be in the following folder structure:
├── data
│ ├── train
│ │ ├── class_name/*.png
...
│ ├── valid
│ │ ├── class_name/*.png
...
│ ├── test
│ │ ├── class_name/*.png
...
└── .gitignore
- Train Use the command:
python3 main.py --dir "data/100-bird-species" --n_classes 5 --b 64 --aug True --epochs 2 --max_lr 0.0001 --n_devices 1 --accelerator "cuda"
To use mlflow go to the project's ROOT directory and use the following command:
mlflow ui
command in the terminal, the API will be allocated on port 5000 of your machine's localhost. Open your browser and type localhost:5000