Datasets:
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/與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