README.md

---
language: en
license: apache-2.0
tags:
- image-classification
- arabic-food
- food-recognition
- vision
- vit
datasets:
- custom
metrics:
- accuracy
model-index:
- name: Arabic Food Classifier
  results:
  - task:
      type: image-classification
      name: Image Classification
    metrics:
    - type: accuracy
      value: 100.0
      name: Test Accuracy
---

# Arabic Food Classifier 🍽️

A Vision Transformer (ViT) model fine-tuned to recognize 10 popular Arabic dishes with **100% test accuracy**.

## Model Description

This model uses Google's ViT-Base architecture, fine-tuned on a custom dataset of Arabic cuisine images. It can accurately identify:

- ☕ Arabic Coffee
- 🫐 Dates  
- 🧆 Falafel
- 🥩 Grilled Meat
- 🫘 Hummus
- 🍚 Kabsa
- 🍰 Kunafa
- 🍛 Mandi
- 🥟 Samboosa
- 🌯 Shawarma

## Performance

**Test Accuracy: 100%** (10/10 correct predictions)

Individual class performance:
- Arabic Coffee: 83.0% confidence
- Dates: 95.7% confidence
- Falafel: 97.4% confidence
- Grilled Meat: 68.7% confidence
- Hummus: 88.5% confidence
- Kabsa: 95.9% confidence
- Kunafa: 98.8% confidence
- Mandi: 53.9% confidence
- Samboosa: 96.5% confidence
- Shawarma: 98.7% confidence

## Usage
```python
from transformers import AutoModelForImageClassification, AutoImageProcessor
from PIL import Image

# Load model
model = AutoModelForImageClassification.from_pretrained("AhmedYasir/arabic-food-classifier-vit")
processor = AutoImageProcessor.from_pretrained("AhmedYasir/arabic-food-classifier-vit")

# Load and process image
image = Image.open("food.jpg")
inputs = processor(image, return_tensors="pt")

# Predict
outputs = model(**inputs)
predicted_class = outputs.logits.argmax(-1).item()

classes = ['arabic_coffee', 'dates', 'falafel', 'grilled_meat', 'hummus', 
           'kabsa', 'kunafa', 'mandi', 'samboosa', 'shawarma']

print(f"Predicted: {classes[predicted_class]}")
```

## Training Details

- **Base Model:** google/vit-base-patch16-224
- **Fine-tuning Method:** Full fine-tuning
- **Dataset:** 992 images (custom Arabic food dataset)
  - Train: 692 images
  - Validation: 144 images  
  - Test: 156 images
- **Training:**
  - Epochs: 3
  - Batch Size: 16
  - Learning Rate: 5e-5
  - Optimizer: AdamW
- **Hardware:** NVIDIA GPU

## Limitations

- Trained on specific Arabic dishes; may not generalize to all regional variations
- Best performance on well-lit, clear images
- Limited to 10 dish categories

## Author

**Ahmed Yasir**
- Building AI/ML systems
- Focus on Arabic language and cultural applications
- [LinkedIn](https://www.linkedin.com/in/ahmed-yasir-907561206/) | [GitHub](https://github.com/ahmedyasir779)

## Citation
```bibtex
@misc{arabic-food-classifier-2025,
  author = {Ahmed Yasir},
  title = {Arabic Food Classifier: Vision Transformer for Arabic Cuisine Recognition},
  year = {2025},
  publisher = {Hugging Face},
  howpublished = {\url{https://huggingface.co/AhmedYasir/arabic-food-classifier-vit}}
}
```

---

Built with ❤️ in Saudi Arabia 🇸🇦