Besides requiring an input shape of 608x608, the saved model gives a much more coarse and less accurate output on the example input: 
Besides requiring an input shape of 608x608, the saved model gives a much more coarse and less accurate output on the example input:
