README.md

---
license: apache-2.0
tags:
- image-classification
- pytorch
- pet-breeds
- oxford-iiit-pet-dataset
- swin-transformer
widget:
  # You might need to configure a Space for the widget to work effectively
  # Or provide a link to your Gradio Space if you create one
  - example_url: "https://huggingface.co/datasets/huggingface/cats-image/resolve/main/cats-300.jpg"
---
# KEDU Breed Classifier (swin_tiny_patch4_window7_224)

This model classifies pet breeds based on the Oxford-IIIT Pet Dataset.
Model architecture: `swin_tiny_patch4_window7_224`.

## How to Use

1.  **Install dependencies:**
    ```bash
    pip install torch torchvision timm pytorch-lightning albumentations opencv-python-headless scikit-learn Pillow huggingface_hub python-box PyYAML
    ```

2.  **Download `inference.py`, `pytorch_model.ckpt`, and `label_encoder.pkl` from this repository.**

3.  **Run inference:**
    ```python
    from inference import load_model_from_hf, load_label_encoder_from_hf, predict_breed

    model = load_model_from_hf(repo_id="Hajorda/keduClassifier")
    label_encoder = load_label_encoder_from_hf(repo_id="Hajorda/keduClassifier")

    image_path = "path/to/your/pet_image.jpg" # Replace with your image path
    predicted_breed, confidence = predict_breed(image_path, model, label_encoder)
    print(f"Predicted: {predicted_breed}, Confidence: {confidence:.4f}")
    ```

## Model Details
- **Dataset:** Oxford-IIIT Pet Dataset
- **Number of Classes:** 37 
- **Image Size:** (224, 224)

## Performance
- **Validation Accuracy:** 93.40%
  (Note: This is the validation accuracy from the training run.)

## Author
Hajorda