FineSKU110k — SKU-110K in the unified detection format

Source: official SKU110K_fixed.tar.gz from the authors' S3 bucket. Repo: https://github.com/eg4000/SKU110K_CVPR19

Converted by the finedet project into a unified, AutoTrain-compatible layout: image / width / height / objects{bbox, category} with COCO-format [x, y, w, h] boxes in absolute pixels. Boxes are clipped to the image and empty boxes dropped; category ids are densified per the category tables below.

Box format

objects.bbox follows the COCO convention: [x, y, w, h] in absolute pixels, origin at the image's top-left corner.

License

Original dataset: SKU-110K (Goldman et al., CVPR 2019), non-commercial license (CC BY-NC). Redistributed non-commercially with attribution.

Example images

Boxes are colored by category: near-transparent fill, opaque outline.

Conversion notes

Single category 'object'. Source boxes are xyxy and converted to COCO xywh. Undecodable images (known corrupt JPEGs in the archive) are skipped and counted.

Splits

• test: 2936 images
• train: 8219 images
• validation: 588 images

Categories

id	name
0	object

Training with transformers

The boxes are already in the absolute-pixel COCO [x, y, w, h] format that AutoImageProcessor expects, so fine-tuning a detector needs no bbox conversion:

import torch
from datasets import load_dataset
from transformers import (AutoImageProcessor, AutoModelForObjectDetection,
                          Trainer, TrainingArguments)

ds = load_dataset("finedet/sku110k")
obj_feat = ds["train"].features["objects"]
if hasattr(obj_feat, "feature"):
    obj_feat = obj_feat.feature
cat_feat = obj_feat["category"]
names = (cat_feat.feature if hasattr(cat_feat, "feature") else cat_feat).names

checkpoint = "facebook/detr-resnet-50"
processor = AutoImageProcessor.from_pretrained(checkpoint)
model = AutoModelForObjectDetection.from_pretrained(
    checkpoint,
    id2label=dict(enumerate(names)),
    label2id={n: i for i, n in enumerate(names)},
    ignore_mismatched_sizes=True,
)


def transform(batch):
    images = [img.convert("RGB") for img in batch["image"]]
    annotations = [
        {"image_id": i,
         "annotations": [
             {"bbox": box, "category_id": cat, "area": box[2] * box[3], "iscrowd": 0}
             for box, cat in zip(objs["bbox"], objs["category"])
         ]}
        for i, objs in enumerate(batch["objects"])
    ]
    return processor(images=images, annotations=annotations, return_tensors="pt")


def collate(batch):
    return {"pixel_values": torch.stack([x["pixel_values"] for x in batch]),
            "labels": [x["labels"] for x in batch]}


trainer = Trainer(
    model=model,
    args=TrainingArguments(output_dir="out", per_device_train_batch_size=4,
                           num_train_epochs=10, learning_rate=1e-5,
                           remove_unused_columns=False),
    train_dataset=ds["train"].with_transform(transform),
    data_collator=collate,
)
trainer.train()