# adobe/antialiased-cnns

pip install antialiased-cnns to improve stability and accuracy

Repository: https://github.com/adobe/antialiased-cnns
Canonical: https://ross.abutalabs.com/products/antialiased-cnns
Homepage: https://richzhang.github.io/antialiased-cnns/
Language: Python
License: NOASSERTION
License Family: other
Topics: convolutional-neural-networks, icml-2019, icml, shift-invariant, shift-equivariant, antialiasing, cnns, artificial-intelligence, computer-vision
Last push: 2024-04-08T12:49:27+00:00

## Health v2 (maintenance only)
Score: 23/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 0, release rhythm 8, longevity 100
- inputs: {"age_days": 2668, "days_push": 877, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_license
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1683, forks 202 (observed 2026-08-28T04:05:21.844945+00:00)

## What it is
A PyTorch library providing antialiased CNN models and a BlurPool layer from the ICML 2019 paper 'Making Convolutional Networks Shift-Invariant Again'. It replaces standard downsampling operations (max-pooling, strided convolution) with low-pass filtered versions to improve shift-invariance, consistency, and accuracy.

## Use cases
- make my CNN robust to small image shifts
- improve classification consistency on ImageNet
- load a pretrained antialiased ResNet as a backbone
- replace max-pooling with a blur pool layer in my architecture
- finetune an existing model with antialiasing
- reduce sensitivity of my vision model to input corruptions

## When to choose
- you need a PyTorch vision backbone with better shift-invariance and robustness
- you want to drop-in replace pooling layers with BlurPool in an existing CNN
- you want pretrained antialiased variants of ResNet, DenseNet, VGG, or similar architectures

## When to avoid
- you are not using PyTorch
- you need transformer-based vision models rather than CNNs
- you need actively developed features or frequent updates

## Facets
- artifact type: library
- maturity: maintenance
- function: machine-learning, image-processing, computer-vision
- domain: deep-learning, computer-vision, machine-learning
- platform: python
- tags: cnn, antialiasing, blurpool, shift-invariance, pytorch, pretrained-models, icml-2019

## Member repositories
- adobe/antialiased-cnns (main) score 23

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:05:21.844945+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 2026-08-30T03:41:18.887046+00:00, confidence not recorded.
  - readme: https://github.com/adobe/antialiased-cnns (fetched 2026-08-28T04:05:21.844945+00:00, sha 6e94dfb2f6b1)
  - homepage: https://richzhang.github.io/antialiased-cnns/ (fetched 2026-08-29T11:14:07.388159+00:00, sha dc84a2612ab7)
  - registry_pypi: https://pypi.org/pypi/antialiased-cnns/json (fetched 2026-08-29T11:14:07.391745+00:00, sha f7e96cdbb8e9)
- Data as of 2026-08-30T08:39:29.467469+00:00.