eye-dataset

Rendered eye images provided by hyunsubs. Upload is in progress; the files currently present are a partial release.

Files

• data/<source-folder>-<index>.tar: image-only, uncompressed TAR archives, each approximately 5 GB or smaller. A source folder's final archive can be smaller.
• manifests/<source-folder>-<index>.jsonl: one record per archived image, with its relative path, bytes, and sha256 checksum.

Image paths inside each TAR preserve their original numbered source folder and file name, for example 01_gray/left_close_1_FinalImage.0000.jpeg. Folders ending in _gray contain grayscale processing outputs. Existing source ZIP archives, model weights, training checkpoints, logs, caches, and experiment directories are excluded from this release. The archives do not define train/validation/test splits or additional labels.

Read an archive

Download an archive from the Files tab, then read its images without extracting the entire archive:

import io
import tarfile
from PIL import Image

with tarfile.open("01_gray-00000.tar", "r:") as archive:
    for member in archive:
        if member.isfile() and member.name.lower().endswith((".jpeg", ".jpg", ".png")):
            with archive.extractfile(member) as stream:
                with Image.open(io.BytesIO(stream.read())) as image:
                    image.load()
                    # Process the image here.

Integrity

Each image is hashed while the archive is built. The completed TAR is read back and checked against the image manifest before upload. Remote TAR size and SHA256 are checked before the temporary local TAR is removed. Original source files are retained locally.

License

CC BY 4.0, as selected by the dataset owner. Attribution: hyunsubs, Eye Dataset, https://huggingface.co/datasets/hyunsubs/eye-dataset.