N

Nat Small In1k 224

Developed by shi-labs
NAT-Small is a hierarchical vision transformer based on neighborhood attention, designed for image classification tasks.
Downloads 6
Release Time : 11/18/2022

Model Overview

NAT is a hierarchical vision transformer based on neighborhood attention (NA), which adopts a restricted self-attention mechanism. The receptive field of each token is limited to its nearest neighboring pixels, with high flexibility and translational equivariance.

Model Features

Neighborhood Attention Mechanism
Adopts a sliding window attention mode, where each token only focuses on its nearest neighboring pixels, achieving local feature extraction while maintaining computational efficiency.
Translational Equivariance
Through the neighborhood attention design, the model maintains equivariance to image translation.
Hierarchical Structure
Adopts a hierarchical vision transformer architecture, suitable for processing visual features at different scales.

Model Capabilities

Image Classification
Visual Feature Extraction

Use Cases

Computer Vision
ImageNet Classification
Classify images into 1,000 categories of ImageNet.
Object Recognition
Identify the main object categories in the image.
Featured Recommended AI Models
AIbase
Empowering the Future, Your AI Solution Knowledge Base
Š 2025AIbase