Skip to content

EfficientNet-B3: Image Classifier

The EfficientNet model provides top-k category predictions out of the 1000 classes on ImageNet based on the EfficientNet B3 algorithm. This network provides state of the art accuracy on ImageNet and at the same time it provides a smaller model and faster inference than other models such as ConvNets. Launch on SageMaker.

The EfficientNet algorithm allows you to also train custom models by fine-tunning (transfer learning) the ImageNet weights to make predictions in custom labels. Launch on SageMaker.


  • State of the art accuracy on the ImageNet validation dataset:
    • Top-1 Accuracy (err): 82.242 (17.758)
    • Top-5 Accuracy (err): 96.114 (3.886)
  • Fine-tunning the ImageNet model to custom labels
  • Provides smaller and faster inference than ConvNets
  • Flexible endpoint to classify one or multiple images with configurable parameters
  • Only pay for what you use with a simple metered pricing model
  • Same price independent of the resources (memory/cpu/gpu) used

Learn how to how to train a custom model, launch this model in AWS SageMaker, see examples from the base model.


This model has a fee of 0.15 dollars per hour. You will incur costs for software use only for as long as the endpoint is running.

AWS infrastructure costs are independent and in addition to the costs of the software and it depends on the instance type selected to host the algorithm.