See our collection for all versions of VGG.

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

zeromodels/vgg16_bn_tv_in1k

Paper: Very Deep Convolutional Networks for Large-Scale Image Recognition (arXiv:1409.1556) · HF Papers

VGG is a classic deep CNN with stacked 3x3 convolutions. Available with and without BatchNorm as ImageNet classifiers / backbones.

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

Pure-Keras 3 conversion of timm/vgg16_bn.tv_in1k for zeromodels. One implementation runs unmodified on TensorFlow / Torch / JAX.

This is an image-classification / backbone checkpoint (VGGImageClassify / VGGModel).

✨ Quick start

import os

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

from PIL import Image
from zeromodels.models.vgg import VGGImageClassify, VGGModel, VGGImageProcessor

model = VGGImageClassify.from_weights("zeromodels/vgg16_bn_tv_in1k")
processor = VGGImageProcessor.from_weights("zeromodels/vgg16_bn_tv_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 = VGGModel.from_weights("zeromodels/vgg16_bn_tv_in1k", as_backbone=True)
features = backbone(pixels, training=False)

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

Tips

• Set KERAS_BACKEND before importing Keras / zeromodels.
• VGGImageClassify returns class logits; VGGModel returns features (as_backbone=True for multi-scale stages).
• See docs and Loading Weights.
• Upstream / timm checkpoints: VGGImageClassify.from_weights("hf:timm/vgg16_bn.tv_in1k").

Special Thanks

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

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