See our collection for all versions of ConvNeXt.

Run ConvNeXt with Keras 3: JAX, PyTorch, or TensorFlow

zeromodels/convnext_large_fb_in22k_ft_in1k_384

Paper: A ConvNet for the 2020s (arXiv:2201.03545) · HF Papers

ConvNeXt modernizes a ResNet-style CNN with ViT-inspired design choices. Use as ImageNet classifier or 4-stage backbone.

For more details on the model, please go to the upstream model card.

Pure-Keras 3 conversion of timm/convnext_large.fb_in22k_ft_in1k_384 for zeromodels. One implementation runs unmodified on TensorFlow / Torch / JAX.

This is an image-classification / backbone checkpoint (ConvNeXtImageClassify / ConvNeXtModel).

✨ Quick start

import os

os.environ["KERAS_BACKEND"] = "torch"  # or "jax" / "tensorflow"

from PIL import Image
from zeromodels.models.convnext import ConvNeXtImageClassify, ConvNeXtModel, ConvNeXtImageProcessor

model = ConvNeXtImageClassify.from_weights("zeromodels/convnext_large_fb_in22k_ft_in1k_384")
processor = ConvNeXtImageProcessor.from_weights("zeromodels/convnext_large_fb_in22k_ft_in1k_384")

image = Image.open("your_image.jpg").convert("RGB")
pixels = processor(image)  # resize + normalize (normalization lives in the processor)
logits = model(pixels, training=False)
print(logits.shape)  # (1, num_classes)

# Feature extraction: the backbone without the classifier head
backbone = ConvNeXtModel.from_weights("zeromodels/convnext_large_fb_in22k_ft_in1k_384", as_backbone=True)
features = backbone(pixels, training=False)

Load any ConvNeXt variant the same way with from_weights("zeromodels/<variant>"):

Variant	Hub
`convnext_atto_d2_in1k`	`zeromodels/convnext_atto_d2_in1k`
`convnext_base_fb_in1k`	`zeromodels/convnext_base_fb_in1k`
`convnext_base_fb_in22k`	`zeromodels/convnext_base_fb_in22k`
`convnext_base_fb_in22k_ft_in1k`	`zeromodels/convnext_base_fb_in22k_ft_in1k`
`convnext_base_fb_in22k_ft_in1k_384`	`zeromodels/convnext_base_fb_in22k_ft_in1k_384`
`convnext_femto_d1_in1k`	`zeromodels/convnext_femto_d1_in1k`
`convnext_large_fb_in1k`	`zeromodels/convnext_large_fb_in1k`
`convnext_large_fb_in22k`	`zeromodels/convnext_large_fb_in22k`
`convnext_large_fb_in22k_ft_in1k`	`zeromodels/convnext_large_fb_in22k_ft_in1k`
`convnext_large_fb_in22k_ft_in1k_384`	`zeromodels/convnext_large_fb_in22k_ft_in1k_384`
`convnext_nano_d1h_in1k`	`zeromodels/convnext_nano_d1h_in1k`
`convnext_nano_in12k_ft_in1k`	`zeromodels/convnext_nano_in12k_ft_in1k`
`convnext_pico_d1_in1k`	`zeromodels/convnext_pico_d1_in1k`
`convnext_small_fb_in1k`	`zeromodels/convnext_small_fb_in1k`
`convnext_small_fb_in22k`	`zeromodels/convnext_small_fb_in22k`
`convnext_small_fb_in22k_ft_in1k`	`zeromodels/convnext_small_fb_in22k_ft_in1k`
`convnext_small_fb_in22k_ft_in1k_384`	`zeromodels/convnext_small_fb_in22k_ft_in1k_384`
`convnext_tiny_fb_in1k`	`zeromodels/convnext_tiny_fb_in1k`
`convnext_tiny_fb_in22k`	`zeromodels/convnext_tiny_fb_in22k`
`convnext_tiny_fb_in22k_ft_in1k`	`zeromodels/convnext_tiny_fb_in22k_ft_in1k`
`convnext_tiny_fb_in22k_ft_in1k_384`	`zeromodels/convnext_tiny_fb_in22k_ft_in1k_384`
`convnext_xlarge_fb_in22k`	`zeromodels/convnext_xlarge_fb_in22k`
`convnext_xlarge_fb_in22k_ft_in1k`	`zeromodels/convnext_xlarge_fb_in22k_ft_in1k`
`convnext_xlarge_fb_in22k_ft_in1k_384`	`zeromodels/convnext_xlarge_fb_in22k_ft_in1k_384`

Tips

Set KERAS_BACKEND before importing Keras / zeromodels.
ConvNeXtImageClassify returns class logits; ConvNeXtModel returns features (as_backbone=True for multi-scale stages).
See docs and Loading Weights.
Upstream / timm checkpoints: ConvNeXtImageClassify.from_weights("hf:timm/convnext_large.fb_in22k_ft_in1k_384").