facebookresearch/hiera

Health v2 · maintenance only

Adoption not part of the score

1074 stars · 61 forks observed · 2026-08-28

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

Hiera is the official PyTorch implementation of a hierarchical vision transformer from Meta AI (ICML 2023 Oral). It achieves state-of-the-art accuracy on image and video tasks while being significantly faster than comparable models by removing unnecessary architectural components.

Use cases

• use a fast vision transformer backbone for image classification
• extract features from images or videos with a pretrained hierarchical transformer
• benchmark vision transformer inference speed against ViT and Swin
• fine-tune a vision transformer for video understanding tasks
• integrate a simple efficient vision backbone into detection or segmentation pipelines

When to choose

• you need a fast, accurate vision transformer backbone for images or video
• you want pretrained models with torch hub or Hugging Face support
• simplicity and inference speed matter more than exotic architectural features

When to avoid

• you need a non-PyTorch framework like TensorFlow or JAX
• you require commercial use of the pretrained model weights (model zoo is CC BY-NC 4.0)
• you need a general-purpose transformer for NLP rather than vision

Facets

library · maturity active

machine-learning deep-learning image-processing video-processing computer-vision deep-learning machine-learning python vision-transformer pytorch image-classification video-understanding backbone facebook-research gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="facebookresearch/hiera")