Kat Tiny Patch16 224.vitft

🚀 Model card for kat_tiny_patch16_224.vitft

KAT model trained on ImageNet-1k (1 million images, 1,000 classes) at resolution 224x224. It was first introduced in the paper Kolmogorov–Arnold Transformer.

🚀 Quick Start

The KAT model is trained on ImageNet - 1k with a resolution of 224x224. It presents a novel approach in the field of image classification.

✨ Features

KAT replaces the channel mixer in transformers with Group Rational Kolmogorov–Arnold Network (GR - KAN), which is a unique design in the model architecture.

📦 Installation

The installation details are not provided in the original README. So, this section is skipped.

💻 Usage Examples

Basic Usage

from urllib.request import urlopen
from PIL import Image
import timm
import torch
import katransformer

img = Image.open(urlopen(
    'https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/beignets-task-guide.png'
))

# Move model to CUDA
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

model = timm.create_model("hf_hub:adamdad/kat_tiny_patch16_224.vitft", pretrained=True)
model = model.to(device)
model = model.eval()

# get model specific transforms (normalization, resize)
data_config = timm.data.resolve_model_data_config(model)
transforms = timm.data.create_transform(**data_config, is_training=False)

output = model(transforms(img).unsqueeze(0).to(device))  # unsqueeze single image into batch of 1

top5_probabilities, top5_class_indices = torch.topk(output.softmax(dim=1) * 100, k=5)
print(top5_probabilities)
print(top5_class_indices)

📚 Documentation

The model definition is at https://github.com/Adamdad/kat, katransformer.py.

📄 License

This model is licensed under the apache - 2.0 license.

Bibtex

@misc{yang2024compositional,
    title={Kolmogorov–Arnold Transformer},
    author={Xingyi Yang and Xinchao Wang},
    year={2024},
    eprint={XXXX},
    archivePrefix={arXiv},
    primaryClass={cs.CV}
}