Skip to content

Author: Danelle Cline,


kclassify runs an image classification model using transfer-learning. This runs on one or many GPUS either locally or in AWS.

Powered by AWS Cloud Computing

Currently, supports the following:


  • Radam
  • Adam
  • Ranger (not working as of 7-20-21)


  • efficientnetB0
  • resnet50
  • vgg16
  • vgg19
  • mobilenetv2


  • width, shift, and zoom
  • horizontal/vertical flip

and all the typical hyperparameters needed for model training like learning rate, batch size, etc.


  • Docker
  • SageMaker SDK version 2.20.0
  • One or more GPUs
  • Training and validation images (JPEG or PNG) images compressed into tar.gz files. See Data organization for details
  • AWS Account (only needed if modifying the model)
    • Your AWS account must support the role ecr:InitiateLayerUpload to push the docker image this creates

Data organization

How to run locally

How to run in AWS