Deeplearning4J Integration - Image Processing
The Image Processing Extension for the KNIME Deeplearning4J Integration allows to use images from KNIME Image Processing as input for deeplearning Nodes. Hence, you can read images, preprocess them using Nodes from KNIME Image Processing and use them to train networks. Images are most commonly used together with convolutional networks.
The Image Processing Extension for the KNIME Deeplearning4J Integration can be found in the KNIP Stable Community Contributions Update Site. By default, this Update Site is not enabled. A tutorial on how to enable the Update Site and how to install an extension can be found here.
Examples Workflows can be found on the public Example Server.
Pixel Type of Images
In order to be used as input, images need to be flattened, which is done automatically. During this process the pixel values will be converted to single precision by the Deeplearning4J Library. In order to problems it is recommended to convert the pixel type of all images to single precision before learning.
The KNIME Deeplearning4J Integration is available under the same License as KNIME.