---
library_name: pytorch
pipeline_tag: image-segmentation
tags:
- pytorch
- image-segmentation
- hair-segmentation
- segface
- korean-hairstyle
- custom-code
language:
- ko
---

# SegFace_k-hair

Personal Hugging Face model repository for a custom PyTorch hair-only segmentation checkpoint trained on a K-Hairstyle based AIHub subset.

## Model Summary

- Backbone: `swin_base`
- Input size: `512x512`
- Freeze backbone: `True`
- LoRA: `rank=8`, `alpha=16.0`, `dropout=0.05`
- Threshold used during validation: `0.5`
- Train / Val split used for this run: `50,000` / `5,000`

## Validation Metrics

These are validation metrics from the training run. A separate independent hold-out test split has not been populated yet, so treat these as validation-only results.

| Metric | Value |
| --- | ---: |
| Best epoch | 7 |
| Val IoU | 0.9487 |
| Val Dice | 0.9736 |
| Val Precision | 0.9723 |
| Val Recall | 0.9751 |
| Epochs completed | 10 |
| Avg epoch time (sec) | 3546.45 |

## Bundle Contents

- `best.pt`: inference checkpoint
- `config.json`: training-time model config
- `training_run_summary.json`: run summary and validation metrics
- `inference.py`: local / Hub inference example
- `requirements.txt`: minimal runtime dependencies
- `hair_mask_dataset/`, `models/`: custom model code required to load the checkpoint

## Inference

Run locally from the root of this model bundle:

```bash
python inference.py
  --checkpoint best.pt
  --input path/to/input.jpg
  --output-mask output_mask.png
  --output-overlay output_overlay.png
```

You can also load directly from the Hugging Face Hub after uploading:

```bash
python inference.py
  --repo-id your-username/SegFace_k-hair
  --input path/to/input.jpg
  --output-mask output_mask.png
  --output-overlay output_overlay.png
```

## Notes

- This repo contains custom code and a raw PyTorch checkpoint, not a Transformers-format model.
- Preprocessing expects RGB input, resize to `512`, ImageNet normalization, and sigmoid threshold `0.5`.
- Before making the repository public, verify whether your AIHub / K-Hairstyle data usage terms allow public redistribution of derived model weights.

## Training Artifacts

![Training Curve](assets/training_curves.png)

![Preview](assets/epoch_010.png)