SegFormer with custom head trained on the People Clothing Segmentation dataset.

This model requires a working CUDA installation. It also uses poetry to manage its dependencies, which can be installed by following the instructions here. The training script also uses the Weights & Biases library, and so you will need to make an account to us the training script as-is (although it does not require much modification to remove the W&B dependency).

Run poetry shell and then poetry install in the project directory before proceeding.

1. Download the Kaggle dataset referenced in the About section to your machine.
2. Run the preprocess.py script, making sure to update the path inside of it to the path to your dataset.

1. Run wandb login.
2. Replace entity="quantumish" to your username and replace the path once more inside main.py.
3. Run main.py.

pdoc is included as a dev dependency. Run pdoc descry to generate documentation.