Which hyper-parameters of a convolutional neural network are likely to be the most sensitive to depending on whether the training (and test and inference) data involves only accurately centered images versus off-centered images.

More convolutional layers, wider convolution kernels, more dense layers, wider dense layers, more or less pooling, or ???

e.g. If I can preprocess the data to include only accurately centered images, which hyper-parameters should I experiment with changing to create a smaller CNN model (for a power and memory constrained inference engine)? Or conversely, if I have a minimized model trained on centered data, which hyper-parameters would I most likely need to increase to get similar loss and accuracy on uncentered (shifted in XY) data?

  • $\begingroup$ Hi and welcome to this community! Maybe you should provide a rigorous definition of "centered images" and "off-centered images" for your question to be answered unambiguously. $\endgroup$ – nbro Jan 1 '20 at 14:12
  • $\begingroup$ Translation augmentation. $\endgroup$ – mirror2image Jan 2 '20 at 8:16

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.