đ ConvNeXT (large-sized model)
A ConvNeXT model pre-trained on ImageNet-22k and fine-tuned on ImageNet-1k at 224x224 resolution, aiming to provide high-performance image classification.
đ Quick Start
ConvNeXT is a pure convolutional model (ConvNet) inspired by Vision Transformers, claiming to outperform them. It was introduced in the paper A ConvNet for the 2020s by Liu et al. and first released in this repository.
Disclaimer: The team releasing ConvNeXT did not write a model card for this model, so this model card has been written by the Hugging Face team.
⨠Features
- Innovative Design: Inspired by Vision Transformers, it modernizes the design of traditional ConvNets.
- High Performance: Claims to outperform Vision Transformers in image classification tasks.
đĻ Installation
No specific installation steps are provided in the original document, so this section is skipped.
đģ Usage Examples
Basic Usage
Here is how to use this model to classify an image of the COCO 2017 dataset into one of the 1,000 ImageNet classes:
from transformers import ConvNextImageProcessor, ConvNextForImageClassification
import torch
from datasets import load_dataset
dataset = load_dataset("huggingface/cats-image")
image = dataset["test"]["image"][0]
processor = ConvNextImageProcessor.from_pretrained("facebook/convnext-large-224-22k-1k")
model = ConvNextForImageClassification.from_pretrained("facebook/convnext-large-224-22k-1k")
inputs = processor(image, return_tensors="pt")
with torch.no_grad():
logits = model(**inputs).logits
predicted_label = logits.argmax(-1).item()
print(model.config.id2label[predicted_label])
For more code examples, we refer to the documentation.
đ Documentation
Model description
ConvNeXT is a pure convolutional model (ConvNet), inspired by the design of Vision Transformers, that claims to outperform them. The authors started from a ResNet and "modernized" its design by taking the Swin Transformer as inspiration.

Intended uses & limitations
You can use the raw model for image classification. See the model hub to look for fine-tuned versions on a task that interests you.
BibTeX entry and citation info
@article{DBLP:journals/corr/abs-2201-03545,
author = {Zhuang Liu and
Hanzi Mao and
Chao{-}Yuan Wu and
Christoph Feichtenhofer and
Trevor Darrell and
Saining Xie},
title = {A ConvNet for the 2020s},
journal = {CoRR},
volume = {abs/2201.03545},
year = {2022},
url = {https://arxiv.org/abs/2201.03545},
eprinttype = {arXiv},
eprint = {2201.03545},
timestamp = {Thu, 20 Jan 2022 14:21:35 +0100},
biburl = {https://dblp.org/rec/journals/corr/abs-2201-03545.bib},
bibsource = {dblp computer science bibliography, https://dblp.org}
}
đ§ Technical Details
ConvNeXT starts from a ResNet and modernizes its design by taking the Swin Transformer as inspiration. It is a pure convolutional model that claims to outperform Vision Transformers in image classification tasks.
đ License
The model is released under the Apache-2.0 license.
Property |
Details |
Model Type |
ConvNeXT (large-sized model) |
Training Data |
ImageNet-22k for pre-training, ImageNet-1k for fine-tuning |
License |
Apache-2.0 |