Request access to MuRI-Restore
Access is manually reviewed by the dataset authors. Please provide accurate information about your identity, affiliation, and intended use.
Access to this dataset requires manual approval. Please provide accurate information about your research purpose and agree to the dataset use terms.
Log in or Sign Up to review the conditions and access this dataset content.
MuRI-Restore Local Parquet Dataset
This dataset was generated from the local data/ directory.
Each top-level folder under data/ is stored as a separate subset/config directory.
License
Dataset files are made available to approved users under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license (CC BY-NC-SA 4.0), with access requiring prior authorization through the Hugging Face gated dataset request workflow.
Public visibility of this dataset card does not grant permission to read, download, copy, or use the dataset files. Users must submit an access request and receive explicit approval from the dataset authors before accessing the files. Approved users may use the dataset only for non-commercial purposes under the CC BY-NC-SA 4.0 terms and the purpose described in their accepted request. Redistribution, mirroring, or public release of the raw dataset files is not permitted without separate written authorization from the dataset authors. Access may be revoked if these terms are violated or if the approved use no longer applies.
Fields
sample_id: stable SHA1-derived id from the relative path.subset: top-level folder name underdata/.image_name: parent folder name forfull.jpg/full.png, otherwise the bitmap filename.hr_image: cropped HR image bytes, PNG encoded. This target image is stored once per row.info_json: raw metadata JSON string ifinfo.jsonexists next to the image.task_spec_json: per-row task metadata and column mapping.- SR columns:
sr_bicubic_x*_lr,sr_natural_x*_lr; image bytes encoded as PNG. - Denoising columns:
denoise_gaussian_sigma*_noisy,denoise_natural_sigma*_noisy; image bytes encoded as PNG. - JPEG columns:
jpeg_q*_compressed; image bytes encoded as JPEG. - Inpainting columns: paired
inpainting_*_maskedandinpainting_*_mask; image bytes encoded as PNG.
Loading Images
Approved users can load this gated Hub dataset directly with
datasets.load_dataset. For gated access, log in first with hf auth login or
provide a valid Hugging Face token in your environment.
The image columns are stored as encoded bytes in Parquet. Decode them with PIL after loading:
from io import BytesIO
from datasets import load_dataset
from PIL import Image
def decode_image(image_bytes):
return Image.open(BytesIO(image_bytes)).convert("RGB")
dataset = load_dataset(
"nomodeset/muri-hf-parquet",
"中国数字博物馆",
split="train",
streaming=True,
)
row = next(iter(dataset))
hr_image = decode_image(row["hr_image"])
lr_image = decode_image(row["sr_bicubic_x2_lr"])
mask_image = Image.open(BytesIO(row["inpainting_random_brush_mask_mask"]))
print(row["sample_id"])
print(hr_image.size, lr_image.size, mask_image.size)
To load the default configuration that combines all subsets:
from io import BytesIO
from datasets import load_dataset
from PIL import Image
dataset = load_dataset("nomodeset/muri-hf-parquet", split="train", streaming=True)
row = next(iter(dataset))
image = Image.open(BytesIO(row["hr_image"])).convert("RGB")
Subsets
上海博物馆_202003: 46 rows中国丝绸博物馆_202003: 309 rows中国南京: 63 rows中国印刷博物馆_20200324: 104 rows中国国家博物馆_202003: 327 rows中国数字博物馆: 52 rows中国美术馆_202003: 99 rows
- Downloads last month
- 4