efficientnet-pytorch

What it is and what it does

EfficientNet PyTorch is a PyTorch reimplementation of the EfficientNet family of convolutional neural networks, which are designed to achieve high image classification accuracy with fewer parameters and lower computational cost than previous models. The package provides pretrained models (EfficientNet-B0 through B8) that can be loaded directly for inference or adapted for transfer learning on custom datasets.

The package supports both standard ImageNet-pretrained weights and adversarially-trained (advprop) variants, with different preprocessing requirements for each. You can use models for classification, feature extraction, or export to ONNX format. However, the package has been abandoned since April 2021 and receives no maintenance, meaning it may not work reliably with modern PyTorch versions or Python environments.

Use it for:

• Load a pretrained EfficientNet model for image classification on new images without retraining.
• Extract learned features from intermediate layers for use in downstream tasks or custom classifiers.
• Fine-tune an EfficientNet on a custom dataset by loading a pretrained model with a different number of output classes.
• Benchmark or compare EfficientNet accuracy against other architectures on ImageNet or custom validation sets.
• Export a trained EfficientNet to ONNX format for deployment in production environments outside PyTorch.

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

Loads and runs pretrained EfficientNet image classification models in PyTorch, supporting both standard and adversarially-trained variants for inference and feature extraction.

No, unless you are locked into an older PyTorch environment. The package is abandoned (last release April 2021, no commits since), so it will not receive security updates, compatibility fixes, or support for modern PyTorch versions. If you need EfficientNet models, consider using PyTorch's built-in torchvision.models or the official TensorFlow implementation instead.

Install

efficientnet-pytorch on PyPI

pip install efficientnet-pytorch

uv add efficientnet-pytorch

poetry add efficientnet-pytorch

Installing efficientnet-pytorch

Before you install

Installation has high friction and the package is abandoned—last release was April 2021, nearly three years ago, with no commits since. While the repository is not archived, the lack of maintenance means security updates and compatibility fixes are unlikely.

License in practice

Licensed under Apache (permissive), so you may use and modify the code freely in commercial or private projects without restriction, though you must retain license notices.

Quickstart

pip install efficientnet-pytorch
from efficientnet_pytorch import EfficientNet
model = EfficientNet.from_pretrained('efficientnet-b0')
# For adversarial training variant:
model = EfficientNet.from_pretrained('efficientnet-b0', advprop=True)

Requires PyTorch to be installed separately; no runtime dependencies listed in package metadata, so you must manage PyTorch installation yourself.

Verify before relying

• Whether the package remains compatible with recent PyTorch versions (last tested in 2021).
• Whether pretrained weights are still accessible from the original sources.
• Whether EfficientNetV2 implementation mentioned in April 2021 update was ever completed and released.

Package facts

Evidence: efficientnet_pytorch-0.7.1.tar.gz

Tags

More Artificial Intelligence packages

Further reading