Model description

This model is based on ViT (Vision Transformer), a transformer-based architecture for image classification that treats image patches as input tokens, enabling the use of pure transformer architectures on vision tasks.

Fine-tuning was done using the transformers Trainer API from Hugging Face.

Intended uses & limitations

You can use this model for:

Classifying breeds of cats and dogs from the Oxford-IIIT Pets dataset.
Fine-tuning on other animal classification datasets.
Serving as a strong vision transformer baseline for academic or benchmarking purposes.

Limitations:

Performance may degrade on images outside of the pet domain.
Not optimized for mobile or edge devices.

Training and evaluation data

Dataset used: pcuenq/oxford-pets

7,390 training images
739 validation images
37 breed classes

Training procedure

Hyperparameters

Learning rate: 5e-05
Train batch size: 32
Eval batch size: 8
Seed: 42
Optimizer: AdamW
Scheduler: Linear
Epochs: 10

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
0.3673	1.0	185	0.3645	0.9080
0.3068	2.0	370	0.3329	0.9161
0.2836	3.0	555	0.3129	0.9175
0.2556	4.0	740	0.2980	0.9202
0.2411	5.0	925	0.2872	0.9215
0.2256	6.0	1110	0.2805	0.9215
0.2378	7.0	1295	0.2751	0.9215
0.2176	8.0	1480	0.2717	0.9215
0.2206	9.0	1665	0.2696	0.9215
0.2173	10.0	1850	0.2690	0.9215

Zero-shot evaluation results

Using a CLIP-based benchmark on the same dataset, the following zero-shot performance was observed:

Accuracy: 0.8800
Precision: 0.8768
Recall: 0.8800

Framework versions

Transformers: 4.50.3
PyTorch: 2.5.1+cu121
Datasets: 3.5.0
Tokenizers: 0.21.1

MaxPowerUnlimited/vit-base-oxford-iiit-pets

vit-base-oxford-iiit-pets