In the model it is required that you must provide 600x800 and 1200x1800 image sizes in order to get a saliency map back. Why is it required that input images must be that size? Those sizes are rather large and contribute to a really slow forward pass time (in my benchmarks I was seeing it take 1.1 seconds on a GPU). Is there any way we can reduce those required sizes?
In the model it is required that you must provide 600x800 and 1200x1800 image sizes in order to get a saliency map back. Why is it required that input images must be that size? Those sizes are rather large and contribute to a really slow forward pass time (in my benchmarks I was seeing it take 1.1 seconds on a GPU). Is there any way we can reduce those required sizes?