ViT Base (Patch16, 224) fine-tuned on Oxford-IIIT Pets 🐶🐱
This model is a fine-tuned version of google/vit-base-patch16-224 on the Oxford-IIIT Pets dataset.
It classifies pet images into 37 cat and dog breeds with high accuracy using transfer learning.
Model Details
- Model type: Vision Transformer (ViT, base, patch size 16, 224 resolution)
- Fine-tuned on: Oxford-IIIT Pets (~37 classes)
- Task: Image classification
- Training framework: 🤗 Transformers
- Approach: Frozen backbone, only classifier head fine-tuned
Training
- Epochs: 5
- Batch size: 16
- Learning rate: 3e-4
- Optimizer: AdamW (default in Trainer)
- Evaluation metric: Accuracy
- Parameters frozen: All except classifier
Evaluation Results
On the test set:
| Metric | Value |
|---|---|
| Accuracy | 93.64% |
| Loss | 0.1834 |
How to Use
from transformers import AutoImageProcessor, ViTForImageClassification
from PIL import Image
import requests
model_name = "ByteMeHarder/basic_vit_transferLearning"
model = ViTForImageClassification.from_pretrained(model_name)
processor = AutoImageProcessor.from_pretrained(model_name)
url = "https://www.robots.ox.ac.uk/~vgg/data/pets/data/images/yorkshire_terrier_1.jpg"
image = Image.open(requests.get(url, stream=True).raw)
inputs = processor(images=image, return_tensors="pt")
outputs = model(**inputs)
preds = outputs.logits.argmax(-1).item()
print("Predicted label:", model.config.id2label[preds])
- Downloads last month
- 2