SHI-Labs/Neighborhood-Attention-Transformer

Health v2 · maintenance only

Flags: no_releases

Adoption not part of the score

1184 stars · 91 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

Official PyTorch implementation of the Neighborhood Attention Transformer (NAT/DiNAT), a family of efficient vision transformers with localized, sliding-window attention, published at CVPR 2023. It provides pretrained models and code for image classification, object detection, segmentation, and image generation tasks.

Use cases

• train a vision transformer for image classification
• semantic segmentation on ADE20K or Cityscapes
• object detection and instance segmentation on COCO
• efficient attention alternative to Swin Transformer
• reproduce CVPR 2023 neighborhood attention paper results
• image generation with StyleNeT

When to choose

• you need an efficient localized-attention vision transformer backbone
• you want pretrained models for detection or segmentation research
• you are reproducing or building on NAT/DiNAT papers

When to avoid

• you need a general-purpose production vision model rather than research code
• you lack a CUDA GPU since the NATTEN extension is required
• you need non-PyTorch frameworks

Facets

library · maturity stable

machine-learning deep-learning image-processing deep-learning computer-vision image-processing python vision-transformer neighborhood-attention pytorch image-classification semantic-segmentation object-detection research-code cvpr gpu linux

Member repositories

For agents

markdown · JSON · MCP: product_card(name="SHI-Labs/Neighborhood-Attention-Transformer")