See our collection for all versions of Res2Net.

Run Res2Net with Keras 3: JAX, PyTorch, or TensorFlow

zeromodels/res2net50_48w_2s_in1k

Paper: Res2Net: A New Multi-scale Backbone Architecture (arXiv:1904.01169) · HF Papers

Res2Net represents multi-scale features at a granular level inside residual blocks. Available as classifier and feature backbone.

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

Pure-Keras 3 conversion of timm/res2net50_48w_2s.in1k for zeromodels. One implementation runs unmodified on TensorFlow / Torch / JAX.

This is an image-classification / backbone checkpoint (Res2NetImageClassify / Res2NetModel).

✨ Quick start

import os

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

from PIL import Image
from zeromodels.models.res2net import Res2NetImageClassify, Res2NetModel, Res2NetImageProcessor

model = Res2NetImageClassify.from_weights("zeromodels/res2net50_48w_2s_in1k")
processor = Res2NetImageProcessor.from_weights("zeromodels/res2net50_48w_2s_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 = Res2NetModel.from_weights("zeromodels/res2net50_48w_2s_in1k", as_backbone=True)
features = backbone(pixels, training=False)

Load any Res2Net variant the same way with from_weights("zeromodels/<variant>"):

Tips

• Set KERAS_BACKEND before importing Keras / zeromodels.
• Res2NetImageClassify returns class logits; Res2NetModel returns features (as_backbone=True for multi-scale stages).
• See docs and Loading Weights.
• Upstream / timm checkpoints: Res2NetImageClassify.from_weights("hf:timm/res2net50_48w_2s.in1k").

Special Thanks

A huge thank you to the Res2Net authors and the timm / Hub communities for creating and releasing these models.

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