deep-tattoo / README.md
jun710's picture
Upload README.md with huggingface_hub
8a0f8a2 verified
|
raw
history blame
940 Bytes
metadata
license: mit
tags:
  - tattoo-segmentation
  - image-segmentation
  - pytorch

Deep Tattoo Segmentation

2-Stage tattoo extraction model using Attention U-Net.

Models

  • best_model_pytorch.pth: Stage 1 - Skin detection (Val Dice: 93.08%)
  • tattoo_stage2_best.pth: Stage 2 v1 - Tattoo extraction (Val Dice: 75.77%)
  • tattoo_stage2_v2_best.pth: Stage 2 v2 - Fine-line tattoo detection (Val Dice: 73.63%)
  • tattoo_final_best.pth: Final model trained on 9,000 images (Best checkpoint)
  • tattoo_final_last.pth: Final model trained on 9,000 images (Last checkpoint)

Usage

import torch
from huggingface_hub import hf_hub_download

# Download model
model_path = hf_hub_download(
    repo_id="jun710/deep-tattoo",
    filename="best_model_pytorch.pth"
)

# Load model
model = torch.load(model_path, map_location='cpu')

Repository

https://github.com/enjius/deep-tattoo