VAN-Tiny

VAN is trained on ImageNet-1k (1 million images, 1,000 classes) at resolution 224x224. It was first introduced in the paper Visual Attention Network and first released in here.

Description

While originally designed for natural language processing (NLP) tasks, the self-attention mechanism has recently taken various computer vision areas by storm. However, the 2D nature of images brings three challenges for applying self-attention in computer vision. (1) Treating images as 1D sequences neglects their 2D structures. (2) The quadratic complexity is too expensive for high-resolution images. (3) It only captures spatial adaptability but ignores channel adaptability. In this paper, we propose a novel large kernel attention (LKA) module to enable self-adaptive and long-range correlations in self-attention while avoiding the above issues. We further introduce a novel neural network based on LKA, namely Visual Attention Network (VAN). While extremely simple and efficient, VAN outperforms the state-of-the-art vision transformers (ViTs) and convolutional neural networks (CNNs) with a large margin in extensive experiments, including image classification, object detection, semantic segmentation, instance segmentation, etc.

Evaluation Results

Model	#Params(M)	GFLOPs	Top1 Acc(%)	Download
VAN-Tiny	4.1	0.9	75.4	Hugging Face 🤗
VAN-Small	13.9	2.5	81.1	Hugging Face 🤗
VAN-Base	26.6	5.0	82.8	Hugging Face 🤗,
VAN-Large	44.8	9.0	83.9	Hugging Face 🤗

BibTeX entry and citation info

@article{guo2022visual,
  title={Visual Attention Network},
  author={Guo, Meng-Hao and Lu, Cheng-Ze and Liu, Zheng-Ning and Cheng, Ming-Ming and Hu, Shi-Min},
  journal={arXiv preprint arXiv:2202.09741},
  year={2022}
}