# sail-sg/poolformer

PoolFormer: MetaFormer Is Actually What You Need for Vision (CVPR 2022 Oral)

Repository: https://github.com/sail-sg/poolformer
Canonical: https://ross.abutalabs.com/products/poolformer
Homepage: https://arxiv.org/abs/2111.11418
Language: Python
License: Apache-2.0
License Family: permissive
Topics: transformer, mlp, pooling, image-classification, pytorch
Last push: 2024-06-01T15:19:56+00:00

## Health v2 (maintenance only)
Score: 23/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 0, release rhythm 8, longevity 100
- inputs: {"age_days": 1745, "days_push": 823, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1362, forks 117 (observed 2026-08-28T04:04:30.345551+00:00)

## What it is
PoolFormer is a PyTorch implementation of the CVPR 2022 paper 'MetaFormer Is Actually What You Need for Vision', which replaces attention in Transformers with simple pooling to show the MetaFormer architecture itself drives performance. It provides pretrained image-classification models achieving 82.1% top-1 accuracy on ImageNet-1K.

## Use cases
- classify images with a lightweight pooling-based vision model
- study whether attention is necessary in vision transformers
- benchmark vision backbones against DeiT and ResMLP
- use a MetaFormer backbone for downstream vision tasks
- reproduce CVPR 2022 PoolFormer results on ImageNet

## When to choose
- you need an efficient vision backbone with fewer parameters and MACs than ViT/MLP models
- you want to experiment with the MetaFormer architecture concept
- you need a simple non-attention token mixer for research

## When to avoid
- you need state-of-the-art attention-based models for production vision tasks
- you need a maintained general-purpose vision framework rather than research code
- you require non-PyTorch frameworks

## Facets
- artifact type: library
- maturity: maintenance
- function: machine-learning, deep-learning, image-processing
- domain: computer-vision, deep-learning, machine-learning
- platform: python
- tags: vision-transformer, pooling, image-classification, metaformer, pytorch, research-code, cvpr-2022, gpu

## Member repositories
- sail-sg/poolformer (main) score 23

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:30.345551+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T04:41:31.648050+00:00, confidence not recorded.
  - readme: https://github.com/sail-sg/poolformer (fetched 2026-08-28T04:04:30.345551+00:00, sha f5e56285ef46)
  - homepage: https://arxiv.org/abs/2111.11418 (fetched 2026-08-29T11:59:16.384449+00:00, sha 2fb79cf3e9db)
  - site_page: https://info.arxiv.org/about/donate.html (fetched 2026-08-29T11:59:16.387620+00:00, sha cca9c3a11c56)
  - site_page: https://info.arxiv.org/about/ourmembers.html (fetched 2026-08-29T11:59:16.391055+00:00, sha 47cbc55ff1de)
  - site_page: https://info.arxiv.org/about (fetched 2026-08-29T11:59:16.392965+00:00, sha a1f16f915a9a)
  - site_page: https://info.arxiv.org/labs/index.html (fetched 2026-08-29T11:59:16.389373+00:00, sha b14a8d05a0ec)
- Data as of 2026-08-30T08:39:29.467469+00:00.