Attempts to run the following two public server workflow examples
- 04_Network_Example_Of_A_MLP_For Images, and
stops at the DL4J Feedforward Learner (deprecated) with the error message "ERROR DL4J Feedforward Learner (deprecated) 2:14:8 Execute failed: No converter for data cell of type Image available."
If I replace the deprecated node with the version 3.3 DL4J Feedforward Learner (Classification) I observe that the "Column Selection" tab in the node configure view does NOT show the presence of the Image column.
Bug or am I doing something wrong?
Characterization of the problem/workaround in KNIME version 3.30:
- When KNIME Image Processing - Deeplearning4J Integration 1.0.2.v201611230738 org.knime.knip.dl4j.feature.feature.group [KNIME Deeplearning4J Integration (64bit only) 3.3.0.v201612012034 org.knime.features.ext.dl4j.feature.group] is installed the node DL4J Feedforward Learner (Classification) works as advertised.
- When "Update KNIME ..." is called, it shows that a new update KNIME Image Processing - Deeplearning4J Integration 1.0.2.v201611230825 has been found
- When that update is installed, the node DL4J Feedforward Learner (Classification) throws the error cited previously
So the workaround is to uninstall the newer version and replace it with the older version
sorry for the late answer. Could it be that you were working with the Nightly Build update site?
Same problem here. Version 3.3 is not working with images. In addition, I can't GPU-enable either version. I get an error in 3.2: Could not initialize class org.nd4j.linalg.factory.Nd4j. I tried both CUDA 7.5 and CUDA 8.
Yes, I was working with the nightly build site, which has the most recent version but is presumable not fully tested. Looks like in the future I will have to work with the latest version release; and not the latest build.
unfortunately we don't guarantee that the nightly build in the community contributions will always work 100%, as it's not considered to be stable and therefore we don't suggest to use it. However, we are of course working on fixing the problem on the nightly build, :-).
Great that it works out for you now!