# facebookresearch/ConvNeXt-V2

Code release for ConvNeXt V2 model

Repository: https://github.com/facebookresearch/ConvNeXt-V2
Canonical: https://ross.abutalabs.com/products/convnext-v2
Language: Python
License: NOASSERTION
License Family: other
Archived: true
Last push: 2024-08-14T15:00:28+00:00

## Health v2 (maintenance only)
Score: 10/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 0, release rhythm 35, longevity 95
- inputs: {"age_days": 1343, "days_push": 749, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, archived, no_license
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 2069, forks 177 (observed 2026-08-28T04:06:10.298916+00:00)

## What it is
Official PyTorch implementation of ConvNeXt V2, a family of pure convolutional neural network models co-designed with a fully convolutional masked autoencoder (FCMAE) framework and a Global Response Normalization (GRN) layer. It provides model definitions in eight sizes, pre-training and fine-tuning code, and pre-trained ImageNet-1K weights.

## Use cases
- classify images with a pretrained ConvNet backbone
- fine-tune ConvNeXt V2 on my own image dataset
- run self-supervised masked autoencoder pre-training on images
- get a lightweight vision model for edge devices
- use a convolutional backbone for downstream vision tasks like detection or segmentation
- compare ConvNet vs Vision Transformer performance on ImageNet

## When to choose
- you need a pure-ConvNets image classification model with pretrained weights
- you want a range of model sizes from tiny (3.7M params) to huge (660M params)
- you want to reproduce or build on the ConvNeXt V2 paper's self-supervised FCMAE approach

## When to avoid
- you need a non-vision or multimodal model
- you require a permissively licensed library since the license is custom/non-standard
- you want a maintained framework with frequent updates, as this is a research code release

## Facets
- artifact type: library
- maturity: stable
- function: machine-learning, deep-learning, image-processing
- domain: computer-vision, deep-learning, machine-learning
- platform: python
- tags: convnext, pytorch, masked-autoencoder, image-classification, self-supervised-learning, pretrained-models, facebook-research, gpu

## Member repositories
- facebookresearch/ConvNeXt-V2 (main) score 10

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:06:10.298916+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 2026-08-30T02:56:45.703390+00:00, confidence not recorded.
  - readme: https://github.com/facebookresearch/ConvNeXt-V2 (fetched 2026-08-28T04:06:10.298916+00:00, sha 110d704337e8)
- Data as of 2026-08-30T08:39:29.467469+00:00.