See our collection for all versions of SENet.

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

kerasformers/seresnet50_a1_in1k

Paper: Squeeze-and-Excitation Networks (arXiv:1709.01507) · HF Papers

SE-ResNet / SE-ResNeXt add Squeeze-and-Excitation channel attention to ResNet/ResNeXt. One package covers both SE-ResNet and SE-ResNeXt Hub variants.

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

Pure-Keras 3 conversion of timm/seresnet50.a1_in1k for kerasformers. One implementation runs unmodified on TensorFlow / Torch / JAX.

This is an image-classification / backbone checkpoint (SENetImageClassify / SENetModel).

✨ Quick start

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

from PIL import Image
import numpy as np
from kerasformers.models.senet import SENetImageClassify, SENetModel

model = SENetImageClassify.from_weights("kerasformers/seresnet50_a1_in1k")
backbone = SENetModel.from_weights(
    "kerasformers/seresnet50_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 SENet variant the same way with from_weights("kerasformers/<variant>"):

Variant	Hub
`seresnet50_a1_in1k`	`kerasformers/seresnet50_a1_in1k`
`seresnext101_32x4d_gluon_in1k`	`kerasformers/seresnext101_32x4d_gluon_in1k`
`seresnext101_32x8d_ah_in1k`	`kerasformers/seresnext101_32x8d_ah_in1k`
`seresnext50_32x4d_gluon_in1k`	`kerasformers/seresnext50_32x4d_gluon_in1k`
`seresnext50_32x4d_racm_in1k`	`kerasformers/seresnext50_32x4d_racm_in1k`

Tips

Set KERAS_BACKEND before importing Keras / kerasformers.
SENetImageClassify returns class logits; SENetModel returns features (as_backbone=True for multi-scale stages).
See docs and Loading Weights.
Upstream / timm checkpoints: SENetImageClassify.from_weights("hf:timm/seresnet50.a1_in1k").