如何更改完全卷积 tflite model 的输入尺寸？

Question

I have a tflite model file for my android application that is created with input dimensions [None,1280,720,3] for RGB input. I would like to use the same model if I also get FHD size image inputs [ None, 1920, 1080, 3] .

Is there a way to change the input dimensions during inference time?

Answer 1

I would recommend training the above model with the new input dimension if you really want to get a better result in terms of accuracy by doing more computations. If not feasible, you can resize the given FHD-sized RGB inputs to the HD sized RGB data as a preprocessing step before invoking the model and then, you can provide the resized data to the same model.