Right now, if the Convolve operator is put on the output_fn of a pattern, as the size of the kernel pattern tends towards zero, the resulting image becomes more and more like the original image. However, when it reaches zero, the result has division errors, instead of becoming identity. It might be necessary to make a special case, perhaps by checking if self.kernel in imagen.sheet_tf.Convolve is all zeros?
Right now, if the Convolve operator is put on the output_fn of a pattern, as the size of the kernel pattern tends towards zero, the resulting image becomes more and more like the original image. However, when it reaches zero, the result has division errors, instead of becoming identity. It might be necessary to make a special case, perhaps by checking if
self.kernel
inimagen.sheet_tf.Convolve
is all zeros?