Darshan764/vit-vehicles-7class

ViT-Base Vehicle Classifier (7 classes)

Fine-tuned google/vit-base-patch16-224-in21k for 7-class vehicle classification.

Classes

Auto Rickshaws, Bikes, Cars, Motorcycles, Planes, Ships, Trains

Results (held-out validation)

Metric	Value
Accuracy	99.46%
Macro F1	99.46%
Train images	4,472
Val images	1,118
Data leakage (train to val)	0

Usage

from transformers import ViTForImageClassification, ViTImageProcessor
from PIL import Image
import torch

model = ViTForImageClassification.from_pretrained('Darshan764/vit-vehicles-7class')
processor = ViTImageProcessor.from_pretrained('google/vit-base-patch16-224-in21k')

img = Image.open('vehicle.jpg').convert('RGB')
inputs = processor(images=img, return_tensors='pt')
with torch.no_grad():
    probs = model(**inputs).logits.softmax(-1)[0]

print(model.config.id2label[probs.argmax().item()], f'{probs.max():.2%}')

Training details

• Base model: google/vit-base-patch16-224-in21k (ImageNet-21k pretrained)
• Epochs: 5
• Batch size: 32
• LR: 2e-5 (cosine schedule, 10 percent warmup)
• Weight decay: 0.05
• Optimizer: AdamW
• Mixed precision (fp16): yes
• Hardware: Colab T4
• Framework: transformers, datasets, PyTorch

How to cite

@misc{Darshan764_vit_vehicles_7class,
  author = {Darshan764},
  title = {ViT-Base Vehicle Classifier},
  year = {2025},
  publisher = {Hugging Face},
  url = {https://huggingface.co/Darshan764/vit-vehicles-7class}
}