1 points | by Johsay 16 hours ago
1 comments
We’re releasing Kanops. Open Acces Imagery (Retail Scenes v0): a gated, evaluation-only dataset of >10,000 in-store photos (UK/US retailers; fixtures, signage, pumpkins/Halloween seasonal, etc.).
HF page: https://huggingface.co/datasets/dresserman/kanops-open-acces...
Shelf/fixture understanding, planogram context, signage OCR, domain adaptation for retail environments. Plus other use cases we are not aware of.
Faces are blurred; metadata & checksums included for reproducibility.
What’s inside Foldered by collection (2014 / FullStores / Halloween2024).
Files: README.md, MANIFEST.csv, metadata.csv, checksums.sha256, LICENSE (eval-only).
Gated access on HF; request and agree to EULA to download.
Quick load (imagefolder) from datasets import load_dataset ds = load_dataset(“imagefolder”, data_dir=“hf://datasets/dresserman/kanops-open-access-imagery/train”) print(ds[“train”][0])
Roadmap v1: add weak/strong labels (CVAT), improved metadata.
Considering a downscaled public “lite” subset; full-res under commercial license.
Happy to answer questions and hear what labels/tasks you’d want first, we have 30 lined up so far,
We’re releasing Kanops. Open Acces Imagery (Retail Scenes v0): a gated, evaluation-only dataset of >10,000 in-store photos (UK/US retailers; fixtures, signage, pumpkins/Halloween seasonal, etc.).
HF page: https://huggingface.co/datasets/dresserman/kanops-open-acces...
Shelf/fixture understanding, planogram context, signage OCR, domain adaptation for retail environments. Plus other use cases we are not aware of.
Faces are blurred; metadata & checksums included for reproducibility.
What’s inside Foldered by collection (2014 / FullStores / Halloween2024).
Files: README.md, MANIFEST.csv, metadata.csv, checksums.sha256, LICENSE (eval-only).
Gated access on HF; request and agree to EULA to download.
Quick load (imagefolder) from datasets import load_dataset ds = load_dataset(“imagefolder”, data_dir=“hf://datasets/dresserman/kanops-open-access-imagery/train”) print(ds[“train”][0])
Roadmap v1: add weak/strong labels (CVAT), improved metadata.
Considering a downscaled public “lite” subset; full-res under commercial license.
Happy to answer questions and hear what labels/tasks you’d want first, we have 30 lined up so far,