Recently I was testing the generalization performance of the RL algorithm, but the size of the images obtained from different environments is inconsistent, which makes it difficult to feed the images into the same neural network.
So, is there a way to make the size of the image returned by the environment consistent?
Question
Recently I was testing the generalization performance of the RL algorithm, but the size of the images obtained from different environments is inconsistent, which makes it difficult to feed the images into the same neural network. So, is there a way to make the size of the image returned by the environment consistent?