Hajorda/keduClassifier

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:

   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:

   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