InkEraser / README.md
ynyg's picture
Update README.md
a057e47 verified
metadata
pretty_name: InkEraser
license: apache-2.0
language:
  - zh
  - en
task_categories:
  - image-to-image
size_categories:
  - 1K<n<10K
tags:
  - image-to-image
  - inpainting
  - handwriting
  - paired-images
configs:
  - config_name: default
    drop_labels: true

ynyg/InkEraser

筆跡擦除數據集,用於將含有筆跡的圖像(source)轉換為去除筆跡的目標圖像(target)。

文件結構

data/
  train/
    source/   # 原圖(含筆跡)
    target/   # 目標(去筆跡)
  val/
    source/
    target/

配對規則

  • source/target/ 中使用完全相同的文件名進行一一對應。
  • 圖像格式:.jpg

數據量(當前倉庫)

  • train:1511 對
  • val:115 對
  • 合計:1626 對

使用示例(Python)

from huggingface_hub import snapshot_download
from pathlib import Path
from PIL import Image

root = Path(snapshot_download("ynyg/InkEraser", repo_type="dataset"))

for split in ["train", "val"]:
    src_dir = root / "data" / split / "source"
    tgt_dir = root / "data" / split / "target"

    for src_path in src_dir.glob("*.jpg"):
        tgt_path = tgt_dir / src_path.name
        # 成對讀取
        src_img = Image.open(src_path)
        tgt_img = Image.open(tgt_path)
        # TODO: your training code

授權

Apache-2.0