zeromodels/resnext50_32x4d_a1_in1k

See our collection for all versions of ResNeXt.

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

zeromodels/resnext50_32x4d_a1_in1k

Paper: Aggregated Residual Transformations for Deep Neural Networks (arXiv:1611.05431) · HF Papers

ResNeXt aggregates residual transformations with cardinality (grouped convolutions). Drop-in ImageNet classifier or multi-scale backbone.

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

Pure-Keras 3 conversion of timm/resnext50_32x4d.a1_in1k for zeromodels. One implementation runs unmodified on TensorFlow / Torch / JAX.

This is an image-classification / backbone checkpoint (ResNeXtImageClassify / ResNeXtModel).

✨ Quick start

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

from PIL import Image
import numpy as np
from zeromodels.models.resnext import ResNeXtImageClassify, ResNeXtModel

model = ResNeXtImageClassify.from_weights("zeromodels/resnext50_32x4d_a1_in1k")
backbone = ResNeXtModel.from_weights(
    "zeromodels/resnext50_32x4d_a1_in1k", as_backbone=True
)

image = Image.open("your_image.jpg").convert("RGB")
image = image.resize((224, 224))
x = np.asarray(image, dtype="float32")[None]  # (1, H, W, 3)
print(model(x).shape)  # (1, num_classes)
feats = backbone(x)
print(len(feats), [tuple(f.shape) for f in feats])

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

Variant	Hub
resnext101_32x16d_fb_ssl_yfcc100m_ft_in1k	zeromodels/resnext101_32x16d_fb_ssl_yfcc100m_ft_in1k
resnext101_32x16d_fb_swsl_ig1b_ft_in1k	zeromodels/resnext101_32x16d_fb_swsl_ig1b_ft_in1k
resnext101_32x16d_fb_wsl_ig1b_ft_in1k	zeromodels/resnext101_32x16d_fb_wsl_ig1b_ft_in1k
resnext101_32x32d_fb_wsl_ig1b_ft_in1k	zeromodels/resnext101_32x32d_fb_wsl_ig1b_ft_in1k
resnext101_32x4d_fb_ssl_yfcc100m_ft_in1k	zeromodels/resnext101_32x4d_fb_ssl_yfcc100m_ft_in1k
resnext101_32x4d_fb_swsl_ig1b_ft_in1k	zeromodels/resnext101_32x4d_fb_swsl_ig1b_ft_in1k
resnext101_32x4d_gluon_in1k	zeromodels/resnext101_32x4d_gluon_in1k
resnext101_32x8d_fb_ssl_yfcc100m_ft_in1k	zeromodels/resnext101_32x8d_fb_ssl_yfcc100m_ft_in1k
resnext101_32x8d_fb_swsl_ig1b_ft_in1k	zeromodels/resnext101_32x8d_fb_swsl_ig1b_ft_in1k
resnext101_32x8d_fb_wsl_ig1b_ft_in1k	zeromodels/resnext101_32x8d_fb_wsl_ig1b_ft_in1k
resnext101_32x8d_tv_in1k	zeromodels/resnext101_32x8d_tv_in1k
resnext50_32x4d_a1_in1k	zeromodels/resnext50_32x4d_a1_in1k
resnext50_32x4d_gluon_in1k	zeromodels/resnext50_32x4d_gluon_in1k
resnext50_32x4d_tv_in1k	zeromodels/resnext50_32x4d_tv_in1k

Tips

• Set KERAS_BACKEND before importing Keras / zeromodels.
• ResNeXtImageClassify returns class logits; ResNeXtModel returns features (as_backbone=True for multi-scale stages).
• See docs and Loading Weights.
• Upstream / timm checkpoints: ResNeXtImageClassify.from_weights("hf:timm/resnext50_32x4d.a1_in1k").

Special Thanks

A huge thank you to the ResNeXt authors and the timm / Hub communities for creating and releasing these models.

License: see YAML license (usually matches the upstream checkpoint).