# wpeebles/gangealing

Official PyTorch Implementation of "GAN-Supervised Dense Visual Alignment" (CVPR 2022 Oral, Best Paper Finalist)

Repository: https://github.com/wpeebles/gangealing
Canonical: https://ross.abutalabs.com/products/gangealing
Homepage: https://www.wpeebles.com/gangealing
Language: Python
License: BSD-2-Clause
License Family: permissive
Topics: cvpr, cvpr2022, gan, image-manipulation, deep-learning, pytorch, computer-vision, alignment, stylegan, stylegan2, tracking, correspondence, generative
Last push: 2022-10-12T22:50:29+00:00

## Health v2 (maintenance only)
Score: 32/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 0, release rhythm 35, longevity 100
- inputs: {"age_days": 1728, "days_push": 1421, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1012, forks 121 (observed 2026-08-28T04:03:13.434793+00:00)

## What it is
Official PyTorch implementation of GANgealing, a CVPR 2022 method that trains a Spatial Transformer to densely align images using GAN-generated training data. It includes pre-trained models, training/evaluation code, and visualization tools for dense visual alignment and correspondence on real images and videos.

## Use cases
- learn dense visual correspondence without labeled data
- align unaligned image datasets to a common template
- propagate edits like cartoon eyes from a template image to real videos
- run mixed reality effects on cat videos with pre-trained spatial transformers
- train a spatial transformer on StyleGAN2 samples
- evaluate self-supervised correspondence algorithms
- warp images with anti-aliased grid sampling and CUDA splatting

## When to choose
- you need self-supervised dense correspondence or alignment without annotation
- you want to propagate annotations or edits from a template to real images/videos
- you are reproducing or building on the GAN-Supervised Dense Visual Alignment paper
- you have GPU resources and want pre-trained models for eight datasets

## When to avoid
- you need a production-ready, actively maintained computer vision library
- you lack a CUDA GPU since key components require it
- your task is generic object detection or segmentation rather than dense alignment
- you need correspondence for arbitrary object categories not covered by the pre-trained GANs

## Facets
- artifact type: library
- maturity: maintenance
- function: machine-learning, deep-learning, computer-vision, image-processing
- domain: computer-vision, deep-learning, machine-learning, image-processing
- platform: python, cross-platform
- tags: pytorch, gan, stylegan2, dense-correspondence, spatial-transformer, visual-alignment, tracking, congealing, research-code, cvpr2022, gpu, linux

## Member repositories
- wpeebles/gangealing (main) score 32

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:03:13.434793+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 2026-08-30T07:11:47.941370+00:00, confidence not recorded.
  - readme: https://github.com/wpeebles/gangealing (fetched 2026-08-28T04:03:13.434793+00:00, sha e77df7c262d5)
  - homepage: https://www.wpeebles.com/gangealing (fetched 2026-08-29T13:11:32.782036+00:00, sha 8bf8c6a84a4b)
- Data as of 2026-08-30T08:39:29.467469+00:00.