This repo provides the code that I used to train the model mattmdjaga/segformer_b2_clothes on hugging face using the mattmdjaga/human_parsing_dataset dataset which is a copy of the ATR dataset for clothes segmentation but can also be used for human segmentation.
Requirements:
- PyTorch
- Transformers
- Evaluate
- Albumentations
- Wandb
- Datasets
For training I prefer using pure PyTorch over the Hugging Face Trainer API. Mainly because I like to have more control over the training loop and I like to use the Wandb library for logging.