SKU-110K: Dense Retail-Shelf Object Detection (No Data Mirror -- Adapter Only)

This card describes a DetectionBench dataset adapter for SKU-110K. It does NOT host or redistribute the dataset itself -- SKU-110K is licensed for the recipient's own academic/non-commercial use, with no redistribution grant. See Getting the Data for the official download link.

Disclaimer

DetectionBench is not an official release of SKU-110K and does not host any SKU-110K images, annotations, or derived files anywhere -- not on Hugging Face, not in this repository.

SKU-110K was created by Eran Goldman, Roei Herzig, Aviv Eisenschtat, Jacob Goldberger, and Tal Hassner at Trax Retail, who retain all rights. This repository does not claim ownership of any images or annotations.

What this repository provides instead:

1. A DetectionBench dataset adapter that converts an official SKU-110K download into DetectionBench's canonical training layout, once you have obtained the data yourself.
2. A helper command, detectionbench-download-dataset --dataset sku110k, that prints (and can fetch) the official download link -- see Getting the Data.
3. This banner, generated locally from a SKU-110K copy already converted through the adapter, purely to illustrate the dataset's domain and annotation style.

---

Dataset Overview

SKU-110K is a dense, single-class retail-shelf object-detection benchmark: 11,743 images of store shelves (8,219 train / 588 val / 2,936 test), with items packed edge to edge -- a stress test for detectors on extreme object density and overlap, rather than fine-grained SKU classification. Despite the "110K" in the name (the number of distinct SKUs pictured across the dataset), the detection task itself is single-class: every box is labeled object.

---

Getting the Data

SKU-110K is not mirrored here. Get it directly from the authors:

detectionbench-download-dataset --dataset sku110k

Resource	Link
SKU-110K dataset (images + CSV annotations)	http://trax-geometry.s3.amazonaws.com/cvpr_challenge/SKU110K_fixed.tar.gz

This is a direct S3 URL -- no Google Drive/Baidu confirmation flow, so detectionbench-download-dataset fetches it automatically with no extra dependencies. Verified against a real download: 8,219 train / 588 val / 2,936 test images (exact match to the official split), 1,208,482 / 90,968 / 431,546 boxes respectively.

Once downloaded, convert it into DetectionBench's canonical layout:

detectionbench-prepare-coco --dataset sku110k --raw-dir <extracted_dir> --output-dir <coco_out>
detectionbench-convert-coco-to-yolo --input-dir <coco_out> --output-dir <yolo_out>

The official annotations/annotations_{train,val,test}.csv split is used as-is.

Classes (1)

object -- single-class, extreme-density detection.

---

Dataset Sources

Original Paper

Precise Detection in Densely Packed Scenes

Eran Goldman, Roei Herzig, Aviv Eisenschtat, Jacob Goldberger, Tal Hassner

IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2019.

Official Resources

• GitHub Repository: https://github.com/eg4000/SKU110K_CVPR19

---

License

"Provided for the exclusive use by the recipient and solely for academic and non-commercial purposes" -- per the official repository. This is more restrictive than a generic non-commercial clause; it does not contemplate onward sharing at all, only the recipient's own use.

Accordingly:

• No Hugging Face mirror of the data is provided or planned.
• The DetectionBench adapter is provided for local, non-commercial research use against a copy you download yourself.
• If you need broader rights, contact the original authors (Trax Retail).

---

Citation

If you use this dataset, please cite:

@inproceedings{goldman2019dense,
  title={Precise Detection in Densely Packed Scenes},
  author={Goldman, Eran and Herzig, Roei and Eisenschtat, Aviv and Goldberger, Jacob and Hassner, Tal},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={5227--5236},
  year={2019}
}

---

Acknowledgements

We sincerely thank Eran Goldman, Roei Herzig, and their co-authors for creating and publicly releasing this valuable dense-detection benchmark.