CadburyGemsYOLODataset

Dataset Description

This dataset contains annotated images for an Automated Optical Inspection (AOI) object detection task, used to identify defective vs. non-defective tablets/candies on a production-line style setup. It was created as a proxy dataset (using Cadbury Gems as stand-ins) for a tablet defect detection use case, and is formatted for training YOLO-family object detectors (YOLOv8, YOLO11, etc.).

• Task: Object Detection
• Classes: 2 (tablet, damaged_tablet)
• Format: YOLO (images + .txt label files with normalized bounding boxes)
• Total images: 796
  • Train: 636 images
  • Validation: 160 images

Dataset Structure

CadburyGemsYOLODataset/
├── train/
│   ├── images/
│   └── labels/
├── val/
│   ├── images/
│   └── labels/
├── classes.txt
└── data.yaml (optional, add if present)

Each image in images/ has a corresponding .txt file in labels/ with the same filename, containing YOLO-format annotations:

<class_id> <x_center> <y_center> <width> <height>

All coordinates are normalized to [0, 1] relative to image dimensions.

Classes

Class ID	Class Name	Description
0	tablet	Intact / non-defective item
1	damaged_tablet	Damaged / defective item

(Defined in classes.txt, one class name per line, in index order.)

Usage

Loading with ultralytics

from ultralytics import YOLO

model = YOLO("yolo11s.pt")
model.train(data="data.yaml", epochs=100, imgsz=640)

Example data.yaml:

path: ./CadburyGemsYOLODataset
train: train/images
val: val/images

names:
  0: tablet
  1: damaged_tablet

Downloading via huggingface_hub

from huggingface_hub import snapshot_download

snapshot_download(
    repo_id="AdityaManojShinde/CadburyGemsYOLODataset",
    repo_type="dataset",
    local_dir="./CadburyGemsYOLODataset"
)

Dataset Creation

Images were collected and annotated using Label Studio, then exported to YOLO format and split into train/validation sets. This dataset was developed as part of an AOI (Automated Optical Inspection) defect-detection pipeline project.

Licensing

Released under the MIT License. Please cite or credit this repository if used in derivative work.

Citation

@misc{cadburygemsyolodataset,
  author = {Aditya Manoj Shinde},
  title = {CadburyGemsYOLODataset},
  year = {2026},
  publisher = {Hugging Face},
  howpublished = {\url{https://huggingface.co/datasets/AdityaManojShinde/CadburyGemsYOLODataset}}
}