vit-base-oxford-iiit-pets

This model was trained to classify cats and dogs and define it's breed using transfer learning method. It 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: 0.2068
• Accuracy: 0.9350

Model description

Since google/vit-base-patch16-224 was used as the base model, the final classification layer was modified to predict 37 classes of cats and dogs from the dataset.

Intended uses & limitations

This model is designed for educational purposes, enabling the classification of cats and dogs and the identification of their breeds. It currently supports 37 distinct breeds, offering a starting point for various learning and experimentation scenarios. Beyond its educational use, the model can serve as a foundation for further development, such as expanding its classification capabilities to include additional breeds, other animal species, or even entirely different tasks. With fine-tuning, this model could be adapted to broader applications in animal recognition, wildlife monitoring, and pet identification systems.

Training and evaluation data

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: 10

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
0.3625	1.0	370	0.2933	0.9269
0.2002	2.0	740	0.2221	0.9432
0.1511	3.0	1110	0.2057	0.9418
0.1253	4.0	1480	0.1876	0.9418
0.1236	5.0	1850	0.1825	0.9432
0.1078	6.0	2220	0.1785	0.9418
0.078	7.0	2590	0.1809	0.9364
0.0798	8.0	2960	0.1785	0.9378
0.0811	9.0	3330	0.1774	0.9364
0.0736	10.0	3700	0.1769	0.9391

Evaluation results

Metric	Value
Evaluation Loss	0.2202
Evaluation Accuracy	92.56%
Evaluation Runtime (s)	7.39
Samples Per Second	100.04
Steps Per Second	12.59
Epoch	10

Framework versions

• Transformers 4.46.3
• Pytorch 2.0.1+cu117
• Datasets 3.1.0
• Tokenizers 0.20.3