I trained a fully convolutional network to transform an image into another image of the same resolution. i was able to convert it into a .tflite model and load it in my android app. I create an interpreter and can set the inputsize by calling
interpreter.resizeInput(0, new int[] {1, image_size, image_size, 3});
but when i use interpreter.run(input,output); it breaks with the message
Cannot copy from a TensorFlowLite tensor (StatefulPartitionedCall:0) with shape [1, 1, 1, 3] to a Java object with shape [1, 1024, 1024, 3].
the output shape should be [1, 1024, 1024, 3] but it isn't.
interpreter.getOutputTensor(0).shapeSignature() and
interpreter.getInputTensor(0).shapeSignature()
return [-1, -1, -1, 3] but i can't change the actual shape you get from
interpreter.getOutputTensor(0).shape()
the way i did with the input by using interpreter.resizeInput(0, new int[] {1, image_size, image_size, 3});
is there a way to implement a tflitemodel that transformes images of variable resolution into other images of the same resolution somehow? I am pretty sure it works before i convert it to tflite
