vit-base-oxford-iiit-pets

This model is a fine-tuned version of google/vit-base-patch16-224 on the pcuenq/oxford-pets dataset. It achieves the following results on the evaluation set:

• Loss: 0.1924
• Accuracy: 0.9445

Model description

This model is a fine-tuned version of a pre-trained Vision Transformer (google/vit-base-patch16-224) for image classification on the Oxford-IIIT Pet Dataset. It uses transfer learning to adapt a generic vision model to identify 37 different cat and dog breeds. The model head is adjusted to output the number of classes in the dataset, and it is trained end-to-end using standard classification loss.

---

Intended uses & limitations

Intended Uses:

• Educational demos on transfer learning and fine-tuning vision models.
• Pet breed classification in structured datasets similar to Oxford Pets.
• Comparative analysis with zero-shot models like CLIP.

Limitations:

• May not generalize well to breeds outside of the Oxford-IIIT dataset.
• Not suitable for real-world medical or safety-critical applications.
• Input images should be clear, centered, and close in style to the training data (cropped pet portraits).

Training and evaluation data

The model is trained and evaluated on the Oxford-IIIT Pet Dataset, which contains 7,349 images of cats and dogs spanning 37 different breeds. The dataset includes equal representation of pets and was split into training, validation, and test sets. Evaluation metrics used include accuracy, precision, and recall.

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0003
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• num_epochs: 5

Training results

Zero-Shot Classification Evaluation (CLIP)

Evaluated the Oxford-IIIT Pet dataset using a zero-shot image classification model: openai/clip-vit-base-patch32. Instead of training, the CLIP model was evaluated using a list of breed names (e.g., "Siamese", "Persian", "Chihuahua") as candidate labels for zero-shot classification.

Evaluation Results:

• Accuracy: 0.8800
• Precision: 0.8768
• Recall: 0.8800

Framework versions

• Transformers 4.50.0
• Pytorch 2.6.0+cu124
• Datasets 3.4.1
• Tokenizers 0.21.1