I know someone had this issue before because their image size is too small to fit in a filter.
In my case, I can use the exact same model in python without any issue (Python 3.8 & Tensoflow 2.2.0 & Keras 2.4.3. But I see the error message when import into DL4J. Is there any version difference to cause the issue? This model contains several dilated convolution with a large dilation factor, The implementation is the same Conv3D with parameter dilate=1,2,4,8, etc…
org.deeplearning4j.exception.DL4JInvalidConfigException: Invalid configuration for layer (idx=-1, name=conv3d_16, type=Convolution3DLayer) for height dimension: Invalid input configuration for kernel height. Require 0 < kH <= inHeight + 2*padH; got (kH=17, inHeight=16, padH=0)
Input type = InputTypeConvolutional3D(format=NCDHW,d=16,h=16,w=16,c=128), kernel = [3, 3, 3], strides = [1, 1, 1], padding = [0, 0, 0], layer size (output channels) = 128, convolution mode = Same