shelfwise-by-form/SKUs_on_shelves_PL

Images of store shelves captured in Poland, with annotated SKUs/products, in COCO format. Creation of this dataset was co-financed from European Funds as part of an R&D project "Recognition of store products using image analysis".

Recommended to use pycocotools to browse the dataset.

The dataset contains:

• ~27k images...
• ...containing ~2M tags...
• ...of ~8k unique SKUs.

Additional comments:

• Photos were taken in stores in Poland and in a laboratory setting (as in the example above).
• All annotations are bounding boxes, but some products were originally annotated with a segmentation mask. Whenever the segmentation field is non-empty it means the bounding box was fitted on the segmentation mask.
• There are 2 auxiliary annotations, both of N/A supercategory:
  • unreadable means a product was annotated with a bounding box, but its category/class couldn't be unambigously determined.
  • unknown_product represents a SKU that was not in the product database at the time of tagging.