BOCCHI Dataset

Blurred Objects Captured across Cameras with Human-annotated Imagery.

633 real-captured images with pixel-level human motion-blur annotations, shot across 5 consumer-grade cameras (SONY α6400 / Nikon D60 / Olympus E-M5 / Canon EOS R8 / Fujifilm X-T30).

Contents

• BOCCHI_dataset/img/0001.jpg … 0633.jpg — RGB images (1080 × 720)
• BOCCHI_dataset/all_data.pkl — pre-processed features + masks bundled together

Load

from huggingface_hub import hf_hub_download
import pickle

pkl = hf_hub_download(
    repo_id="BrianChen54923/BOCCHI",
    filename="BOCCHI_dataset/all_data.pkl",
    repo_type="dataset",
)
data = pickle.load(open(pkl, "rb"))

Or for individual images, use HfApi().hf_hub_download() on any img/NNNN.jpg.

Mask Convention

white = sharp, black = blur.

License

CC BY-NC 4.0 — research use only.