zeromodels/resmlp_big_24_224_fb_in1k

See our collection for all versions of ResMLP.

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

zeromodels/resmlp_big_24_224_fb_in1k

Paper: ResMLP: Feedforward networks for image classification with data-efficient training (arXiv:2105.03404) · HF Papers

ResMLP is a residual MLP architecture for vision with data-efficient training. Classifier or block features.

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

Pure-Keras 3 conversion of timm/resmlp_big_24_224.fb_in1k for zeromodels. One implementation runs unmodified on TensorFlow / Torch / JAX.

This is an image-classification / backbone checkpoint (ResMLPImageClassify / ResMLPModel).

✨ Quick start

import os

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

from PIL import Image
from zeromodels.models.resmlp import ResMLPImageClassify, ResMLPModel, ResMLPImageProcessor

model = ResMLPImageClassify.from_weights("zeromodels/resmlp_big_24_224_fb_in1k")
processor = ResMLPImageProcessor.from_weights("zeromodels/resmlp_big_24_224_fb_in1k")

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 = ResMLPModel.from_weights("zeromodels/resmlp_big_24_224_fb_in1k", as_backbone=True)
features = backbone(pixels, training=False)

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

Variant	Hub
resmlp_12_224_fb_distilled_in1k	zeromodels/resmlp_12_224_fb_distilled_in1k
resmlp_12_224_fb_in1k	zeromodels/resmlp_12_224_fb_in1k
resmlp_24_224_fb_distilled_in1k	zeromodels/resmlp_24_224_fb_distilled_in1k
resmlp_24_224_fb_in1k	zeromodels/resmlp_24_224_fb_in1k
resmlp_36_224_fb_distilled_in1k	zeromodels/resmlp_36_224_fb_distilled_in1k
resmlp_36_224_fb_in1k	zeromodels/resmlp_36_224_fb_in1k
resmlp_big_24_224_fb_distilled_in1k	zeromodels/resmlp_big_24_224_fb_distilled_in1k
resmlp_big_24_224_fb_in1k	zeromodels/resmlp_big_24_224_fb_in1k
resmlp_big_24_224_fb_in22k_ft_in1k	zeromodels/resmlp_big_24_224_fb_in22k_ft_in1k

Tips

• Set KERAS_BACKEND before importing Keras / zeromodels.
• ResMLPImageClassify returns class logits; ResMLPModel returns features (as_backbone=True for multi-scale stages).
• See docs and Loading Weights.
• Upstream / timm checkpoints: ResMLPImageClassify.from_weights("hf:timm/resmlp_big_24_224.fb_in1k").

Special Thanks

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

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