---
license: apache-2.0
task_categories:
- image-classification
language:
- en
tags:
- fine-grained-classification
- agricultural-vision
- adulteration-detection
- food-quality
- saffron
size_categories:
- n<1K
---

# SaffronVerify Dataset

## Dataset Summary
SaffronVerify is a fine-grained image classification dataset for saffron quality grading. It contains images of saffron across three quality categories — from premium grade to adulterated samples — intended for training computer vision models to detect saffron purity and adulteration.

## Dataset Structure
The dataset follows the standard ImageFolder layout and is split into training and validation sets.

```text
saffron-verify/
├── train/
│   ├── mogra/
│   ├── lacha/
│   └── adulterated/
└── val/
    ├── mogra/
    ├── lacha/
    └── adulterated/
```

### Splits
| Split | Description             |
|-------|-------------------------|
| train | 80% of total images     |
| val   | 20% of total images     |

## Classes
| Label       | Original Class | Description                                              |
|-------------|----------------|----------------------------------------------------------|
| mogra       | class A        | Premium grade saffron — deep red, fully intact stigmas   |
| lacha       | class B        | Mixed grade — contains yellow styles along with red stigmas |
| adulterated | class C        | Adulterated saffron — broken threads, debris, foreign matter |

## Image Format
- All images are saved as JPEG (`.jpg`) at quality 95  
- All images are converted to RGB (3-channel)  
- Input sources included `.jpg`, `.png`, and `.webp` originals

## Loading the Dataset
```python
from datasets import load_dataset

ds = load_dataset("Arko007/saffron-verify")
print(ds)
# DatasetDict({
#     train: Dataset({features: ['image', 'label'], num_rows: ...})
#     val:   Dataset({features: ['image', 'label'], num_rows: ...})
# })

print(ds["train"].features["label"].names)
# ['adulterated', 'lacha', 'mogra']
```
## Intended Use
This dataset is intended for:
- Training image classifiers for saffron quality grading
- Research on agricultural product adulteration detection
- Benchmarking fine-grained food quality classification models

## Preprocessing
- Dataset was preprocessed using a custom Python pipeline  
- Random train/val split with seed 42 for reproducibility  
- Images sequentially renamed per class per split (e.g. `mogra_001.jpg`)  
- RGBA images composited to RGB prior to saving

## License
This dataset is released under the MIT License.

## Citation
If you use this dataset in your work, please cite:

```text
@dataset{saffronverify2026,
  author    = {Arko007},
  title     = {SaffronVerify: Saffron Quality Classification Dataset},
  year      = {2026},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/datasets/Arko007/saffron-verify}
}
```